6sense
Ask an AI agent about 6sense account insights, buying signals and keyword trends.
One MCP server for an Ambiguous workspace: mail, docs, sheets, chat, calendar, CRM, tasks.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
The Ambiguous MCP server gives Claude a whole Ambiguous workspace through one connection: mail, docs, sheets, slides, chat, calendar, drive, tasks, CRM, wiki, forms and e-signature. Our live listing returned 914 tools. Many act outward — sending email, posting chat messages, emailing quotes, inviting calendar attendees — and each runs with your workspace role.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
List users in the active workspace. Powers @mention pickers, assignee dropdowns, etc. Filterable by q (matches display_name + username substrings or an owned workspace email address, case-insensitive). Limit defaults to 50, hard cap 100. Member-accessible — use this to resolve a name to a user_id when assigning tasks or @mentioning people; the admin list_users tool is admin-gated and 403s for members/coworkers.
Return the current authenticated user's profile (works with both JWT and API-key auth), including active workspace name + ID and agent kind + manager when applicable. needs_workspace_setup: true indicates the account has no workspace yet.
List the caller's own active human client sessions across workspaces, current first then last server communication descending. A session is one authentication lineage, not an inferred physical device. Excludes agent credentials, API keys, app passwords, OAuth connections and memberships. Valid legacy sessions join the inventory when first observed; renewal writers need not be enabled.
Sign out one of the caller's own human sessions, including its renewed JWTs, workspace selections and renewal family. Unknown or foreign session ids return404; repeating an own-session revoke succeeds without another effect. Current-session revocation requires the client to use its ordinary local exit flow.
List accessible documents, spreadsheets, presentations and canvases in the current workspace, returning IDs, titles, types and timestamps. Filter by type and paginate with limit/offset. scope=account includes documents owned in other workspaces; scope=workspace also includes documents shared here with the user's other account identities.
Create a new doc. `content` is Markdown (recommended) or a ProseMirror JSON string. Returns the created document; read, update, share and trash it through the `/api/documents/{id}` operations. For a spreadsheet use `create_sheet`, for a deck `create_presentation`, for a whiteboard `create_canvas`.
Get the caller's default layout for new Docs: `pageless_default: true` (the default) means new Docs start pageless, `false` means they start in pages.
Set the caller's default layout for new Docs (`pageless_default: true` = pageless, `false` = pages). Affects only Docs created afterwards; existing Docs keep their layout.
Documents shared with the calling account through an explicit permission grant, across every workspace and every identity the account owns. Excludes documents the account owns. Each row names the workspace that shared it and the role the grant carries. `scope=workspace` narrows discovery to resources in the authenticated current workspace without changing access.
Find all deleted native Docs, Sheets, Slides and Canvas items owned by the caller. Returns their document IDs for documents_restore or documents_permanent_delete; optional type filters doc, sheet, slide or canvas. Uploaded Drive files and folders use list_trashed_files, which can also include materialized native-document shadow rows but is not a complete native-document trash list. An empty Drive-file trash result does not mean this document trash is empty.
Restore a trashed document.
Make a copy. Title becomes 'Copy of <orig>'. Visibility resets to restricted; labels/pin/star are NOT inherited. Requires viewer access on the source.
Export a document, sheet, or slide to one of its supported standard formats. Sheets support xlsx, csv, tsv, and ods. Returns the file bytes with Content-Type and Content-Disposition attachment headers.
Start a PDF or DOCX export of a document, sheet, or slide without waiting for the render. Returns 202 with the job and status_url at once; the Location header points to the same path. Poll status_url until status is succeeded, then fetch download_url, or failed, whose error carries the status the synchronous export would have answered. The same idempotency_key returns the same job. Requires viewer access, and only the requester can read the job.
Read an export job started with documents_export_job_create. When status is succeeded, download_url is a short-lived signed URL for the file; read the job again for a new one. Only the requester, while still able to view the document, can read it; anyone else gets 404.
The finished file of a succeeded export job, as an attachment, for the requester only. download_url names this path when the storage cannot sign a URL; otherwise use the signed URL.
Get a single document by ID, including its full content and comments. If the user is currently viewing a document, use the document ID from the page context.
Edit an existing document inline. Read the document first (get_document), modify the content, then pass the full updated text — content replaces the entire body. Do NOT create a new document to edit; version history preserves prior states. Accepts Markdown or ProseMirror JSON (HTML tolerated but lossy — drops listed in `import_warnings`). Provide only the fields you want to change.
Activate source-preserving editing on an existing document. Only its owner may migrate; the expected content hash prevents stale activation and the command ID makes retries idempotent.
Prepare compatible readers and activate source-preserving paragraph structure editing on a generation-zero document. Requires edit permission, a current content hash and state vector, and a completed reader fence. Pending fence returns its ready-after deadline without activating the model.
List pending suggestions and opaque complete-occurrence targets at the current content hash. Requires owner or editor access.
Accept or reject selected opaque suggestion targets, or all pending suggestions, against the content hash returned by list_document_suggestions. Requires owner or editor access.
Create, edit or remove typed paragraph and character style definitions with a guarded whole-registry command. Read current content first, preserve sparse native declarations and include the current content SHA-256. Explicit adoption authorizes replacing supported imported declarations; projection scope authorizes only derived style map/native carrier changes. Initial existing-document writes return a reader fence deadline; retry the same command after it drains. Concurrent stale changes refuse. The inverse registry is a new command guarded by the latest content hash. An optional restore_projections style-only preimage obeys the document body/walk limits and is accepted only when a forward replay reconstructs the current stored style projections.
Soft-delete a document (move to trash). Owner only.
Content-derived thumbnail for doc/sheet/slide cards. Returns SVG by default or PNG for Accept: image/png; cached privately for 60 seconds.
Add an image to a doc/sheet/deck/board. Returns the parent-owned image URL `/api/assets/{id}/content`. Input: content_base64 bytes, source_file_id (copies a viewable Drive file), or source_url (copies public HTTPS through SSRF guard); no source link retained. Native sources without stored bytes return 422 native_document_not_copyable; attach it as a link. Byte-sniffed images only, 415 for other types or malformed SVG. SVG sanitized on upload and every read, inline image/svg+xml with restrictive CSP; malformed stored SVG returns 404. Drive SVG stays attachment. Owner quota exceeded 413; same image deduplicated per parent. Absent from Drive, readable exactly with parent.
Toggle pinned state. Editor permission required (workspace-wide pin, not per-user).
Get the default, first-page, and even-page headers and footers from a document as ProseMirror JSON nodes (or null if not set).
Set default, first-page, and even-page headers and footers. Each field accepts a plain text string (auto-wrapped in a paragraph), a ProseMirror JSON node, or null to remove. Omit a field to leave unchanged.
Get the footnotes from a document. Returns an array of footnotes, each with a footnoteId, body content (PM JSON), and the positions of inline references (footnoteRef nodes) in the body text.
Set footnote body content on a document. Replaces the footnoteSection with canonical footnoteItem nodes. Each footnote must have a footnoteId and either text, inline PM JSON content, or one paragraph without paragraph formatting. Bodies that cannot be represented losslessly are refused. When stored model requirements apply, send model_features covering them; missing or incomplete required capabilities are refused with 409. To add inline footnote references in the body text, use update_document with PM JSON content containing footnoteRef nodes with matching footnoteIds.
Get the page-level formatting for a document: margins (in PT), page size (letter/A4/legal or custom {width, height} in PT), and orientation (portrait/landscape). Empty object means default formatting.
Set page-level formatting for a document: margins (PT), page size (letter/A4/legal or {width, height} in PT), orientation (portrait/landscape). 72 PT = 1 inch. These settings control export output (.docx, .pdf). Pass an empty object to reset to defaults.
Make an unclaimed document yours. A document, spreadsheet, presentation or board created without an account returns a single-use `claim_key`; call this with it after signing in and the document moves into your active workspace with you as owner, keeping its id and URL. Its link drops to view-only, which you can raise again. A second claim of the same document, or a wrong key, is refused with 409 `claim_rejected`, which does not say which.
List version snapshots, newest-first, cursor-paginated (default 50 per page). Walk via `next_cursor` until `has_more` is false.
Create a manual snapshot (forces past the 5-minute throttle).
Get a version's metadata + serialized content snapshot.
Rename a version snapshot of a document. Useful for labeling meaningful versions (e.g., "Pre-launch draft", "Board approved"). Requires editor permission.
Delete a version snapshot of a document. At least one version must always remain — returns an error if attempting to delete the last version. Requires editor permission.
Revert the document to a prior version. Force-snapshots current state first.
Create a named range (template placeholder) on a document. Named ranges anchor to ProseMirror node positions and are used for template-based document generation. Use replace_named_range_content to fill a named range with content. Name must be alphanumeric + underscore, must not start with a digit (e.g., party_name, companyAddress).
List all named ranges (template placeholders) defined on a document. Returns each range's name and its anchored node positions.
Delete a named range (template placeholder) from a document by name. Names are unique per document.
Replace all content within a named range with new content. Used for template-based document generation — define named ranges as placeholders, then call this to fill each one. The named range must have positional anchors (startNodeId/endNodeId). Content can be Markdown text or PM JSON. Returns 409 if the anchored nodes no longer exist (stale range after CRDT edits).
Insert a page break into a document at a given position. Forces subsequent content to start on the next page in exported .docx and .pdf files. The page break is inserted as a block node between paragraphs.
Insert a section break into a document at a given position. Section breaks divide the document into sections that can have different formatting. A 'next_page' section break forces content to the next page (like a page break but marks a new section). A 'continuous' section break starts a new section on the same page.
Insert a page number placeholder into a document. The placeholder renders as the current page number in exported .docx and .pdf files. Commonly used inside headers or footers (e.g., 'Page X') but can also appear inline in the body text. The placeholder is inserted as an inline node within a paragraph.
Set the number of pinned (repeating) header rows on a table in a document. Pinned header rows repeat on every printed page when the table spans multiple pages in exported .docx and .pdf files. Set to 0 to clear the pinned header rows attribute.
Move the caller's project-less tasks linked from this doc into `project_id`; set the doc's tasks project if unset. Other linked tasks are listed in `skipped` and unchanged.
List the user's chat channels. Returns channels with id, name, type (public/private/dm), description, member count, and creator info. Use this when the user asks to see their channels, find a specific channel by name, or when you need a channel ID to send a message.
Browse the workspace's public channels, including the ones the caller has not joined, each with a `joined` flag and member count. `list_channels` returns memberships only, so this is the only way to discover a public channel you are not in and get the id `chat_channel_join` needs. Private channels and DMs are never listed. Use this when the user asks what channels exist, or asks to join a channel by name.
Create a chat channel in the workspace.
Mark all channels as read for the current user. Advances the read cursor in every channel the user is a member of to the latest message, clearing all unread badges at once. Use this when the user wants to dismiss all chat unreads (e.g. returning from vacation).
Batch-fetch the latest message preview for multiple channels in one request. Pass `ids` as a comma-separated list of channel IDs (max 100). Returns one entry per channel the caller can read, each carrying its `channel_id`. Replaces per-channel `GET /messages?limit=1` calls so a DMs/channels list renders previews without an N+1 request fan-out.
Rank the caller's channels and DMs by their own recent engagement: posts, reactions, and mentions of the caller over the last 30 days, each decaying with a 7-day half-life. Only channels the caller can view; score descending.
Find existing chat messages by keyword, phrase, or topic across the caller's accessible channels. Results include the message id and channel context needed to bookmark, remind, react, reply, or edit that message. Use channel_id to restrict the search to a known channel; this read does not post messages.
Acknowledge readable Threads inbox activity on entry without advancing personal read cursors or clearing explicit manual unread intent.
List thread-activity inbox for the current user. Powers the sidebar "Threads" inbox entry. Returns readable threads the user participates in or explicitly marked unread, with peer-only unread counts and own-inclusive unread anchors.
Acknowledge the Mentions & reactions inbox after it has been displayed. Marks the caller's unread chat mention notifications read and stamps users.mentions_acked_at, so the projector recomputes both chat.mentions and chat.reactions to 0 durably (the sidebar "Mentions & reactions" item un-bolds and stays cleared across re-fetch). Per-channel read cursors are untouched, so channel message unread counts are unaffected; a reaction that arrives after the ack re-bolds the item.
List all messages the current user has saved (bookmarked) across all channels. Returns messages with saved_at timestamp, channel name, and channel type. Saved messages are private to the current user (unlike pinned messages, which are shared with the channel).
List all pending scheduled messages for the current user across all channels. Returns each scheduled message with channel name/type and the scheduled_at timestamp. Use this to review what's queued to send later, or before updating/cancelling a scheduled message.
Edit the content or scheduled send time of a pending scheduled message before it's sent. Fails if the message has already been sent or cancelled. Use this when the user wants to fix a typo in a scheduled message or reschedule it.
Cancel a pending scheduled message so it's never sent. Use this when the user says 'never mind' or changes their plan before the scheduled time.
Deliver a pending scheduled message immediately instead of waiting for its scheduled time. Flips the message to sent and fans it out to the channel right now. Fails if the message has already been sent or cancelled. Use this when the user wants to send a queued message without waiting.
List all custom emojis uploaded to the current workspace. Returns each emoji with id, name (without colons), image URL, and creator info. Use this to discover what custom emojis are available before referring to them by name.
Serve custom emoji image from GCS
Delete a custom emoji from the workspace. Use this when the user asks to remove or clean up a custom emoji. This does not remove existing reactions or messages that used the emoji — it only stops new usage.
Fetch OpenGraph / link-preview metadata for a URL (title, description, image, site name, favicon). Useful for generating rich previews of shared links, or inspecting what a URL would look like when shared in chat. Only http(s) URLs are allowed, and internal/private addresses are blocked.
Get the current user's active chat status (emoji + text) and DND-until timestamp. Use this before setting or clearing the status to know the current state.
Set the current user's chat status — an emoji, short text (<=100 chars), and optional auto-clear timestamp. All fields are optional; pass null to clear a single field. Use this when the user says things like 'set my status to lunch until 1pm' or 'I'm on vacation'.
Clear the current user's status (emoji + text + expiresAt). Does NOT change DND — use set_user_dnd with until=null to clear DND separately.
Enable or disable Do-Not-Disturb for the current user. Pass an ISO-8601 timestamp in the future to snooze notifications until that time, or pass null to disable DND immediately. Use this when the user says 'mute me for an hour', 'DND until 5pm', or 'turn off DND'.
List active chat statuses (emoji + text + DND) for users in the workspace. Only returns users who currently have a non-null status or active DND. Useful for 'who's on vacation?', 'is anyone in meetings?', or similar presence queries.
Get the current user's notification keywords. The user receives a mention-style notification whenever any of these words appears in a message they can read.
Replace the current user's notification keyword list. Keywords are case-insensitive, whole-word, and capped at 20 entries (each up to 50 characters). Duplicates/empties are dropped. Use this when the user says 'alert me whenever someone mentions X' or 'stop notifying me about Y'.
Choose your channel notification level: all_messages, new_threads_mentions, mentions_only, or nothing (Off). A direct message (1:1 or group) has no level: it always behaves as all_messages, mute is its only control, and this write is refused 400 dm_notification_level_unsupported. New human memberships default to all_messages; agent and coworker memberships default to mentions_only. All messages includes every new message and thread reply. New threads + mentions includes new top-level messages, direct mentions, broadcasts, and followed-thread replies. Mentions only retains addressed messages and followed-thread replies. Off suppresses notifications and attention badges while retaining readable content and unread bolding. DND pauses all Chat notifications; temporary mute still admits addressed mentions. The legacy unread_only input maps to all_messages for humans and mentions_only for agents and coworkers. Per-member; other members are unchanged. Pass user_id to set an agent's level as an admin, owner or the agent's manager; nobody sets another human's level.
Self-join a public channel (no body required)
Leave a chat channel you belong to.
Mark a channel as read up to a specific message. This clears the unread badge for the current user. Use this after the user has finished catching up on a channel, or programmatically to dismiss unread counts after the assistant reports on recent messages.
Mark a channel as unread starting from a specific message. Moves the calling user's read cursor back to just before that message, so the channel shows an unread badge and the message becomes the unread boundary. Use this when the user wants to revisit a conversation later.
Get per-member read positions for a channel — who has read up to which message. Useful for reporting on whether an announcement has been seen by everyone.
List all pinned messages in a channel. Pinned messages are highlighted by members for ongoing reference (announcements, canonical links, etc.). Returns messages in newest-pin-first order.
Get saved message IDs for a channel (for UI bookmarked state)
Add member(s) or a group
Remove a member from a chat channel.
Change a channel member's role to 'admin' or 'member'. Requires channel admin privileges. Use this when the user asks to promote or demote a member in a channel.
Remove a group from channel (clean up fan-out rows)
Get recent messages from a chat channel. Returns messages with id, content (Markdown), author info, reactions, thread reply count, and timestamps. Use this when the user asks to read messages, catch up on a channel, or find a specific conversation. If the user is currently viewing a channel, use the channel ID from the page context. Messages are returned in reverse chronological order (newest first).
Post a NEW message visible to the chat channel's members, or send a thread reply. Use for requested communication, not for finding, bookmarking, or reminding about an existing message: those actions use search_chat_messages/get_channel_messages then save_message/set_reminder on the returned message id. Content is Markdown with @mentions, code blocks, and links. Use the current channel ID when supplied by page context. To reply, set thread_id to the original parent message's id. To attach a file, upload it with drive_file_upload_proxy (source=chat) and include [name](drive://<file_id>) in content.
Edit an Ambiguous-origin message or pin/unpin. Slack-origin content must be edited in Slack (409).
Delete one message from a chat channel. Slack-origin messages must be deleted in Slack (409).
Soft-delete up to 100 messages in a channel. Deletes only the IDs the caller may delete (own messages, or any message when the caller is a channel admin); every other ID returns in `rejected`. A batch affecting Slack-origin messages is refused with 409 before any deletion. Emits one bulk `message.deleted` event, never one per message.
List the users who have read a specific message (i.e. whose read-cursor is at or past this message). Use this to check whether a message (e.g. an important announcement) has been seen by specific people.
Bookmark an EXISTING chat message in the current user's private Later/saved list. Resolve the original message_id with search_chat_messages or get_channel_messages when it is not supplied. Does not post or copy a message, and does not set a timed reminder (use set_reminder for that). Saving an already-saved message is a no-op.
Remove a saved (bookmarked) message from the user's saved list. The message itself is untouched; only the personal bookmark is deleted.
Remind only the signed-in user about an existing chat message at a future time. A channel named in a personal reminder request locates the source message; it is not an audience or a request to post there. When message_id is unknown, first find the original with search_chat_messages or get_channel_messages; an unknown ID does not mean the message is absent. If the target cannot be identified, ask for clarification instead of publishing a replacement. Automatically bookmarks the original in Later. remind_at is an ISO 8601 datetime and must be in the future.
Clear a pending reminder on a saved message. The message stays saved; only the reminder is removed.
Re-snooze an existing saved message by updating its remind_at time. Resets reminded to false so the scheduler picks it up again. Accepts ISO 8601 datetime string; must be in the future.
Get the replies in a chat message's thread.
Mark a thread as fully read up to the latest reply. Use this after the user has finished catching up on a thread, or when you want to programmatically dismiss unread reply counts for a specific thread.
Mark a thread as unread so it reappears in the threads inbox with an unread badge. An optional reply_id marks that reply onward unread. Omit it to mark the whole thread unread. Restore accepts the prior cursor and its expected current cursor; a newer intent returns 409.
Read identity-only evidence for your retained keyed message in the current workspace. Requires Chat read and current channel membership. Live and spent both confirm an accepted retained row; spent returns no deleted content. Neither certifies delivery. A missing, purged, foreign, or inaccessible result is the same generic 404/unknown, never proof that sending did not occur and never permission to resend or replace the key. This Read posts nothing and creates no membership.
Messages authored by the caller that carry reactions from other users. Powers the Mentions & Reactions inbox.
Messages from other users that @-mention the caller. Powers the Mentions tab.
Workspace-level chat preferences readable by any authenticated member (default_reactions).
Get whether preview cards render under chat messages for the caller (`show_link_previews: true`, the default). The setting is the caller's own; it never changes what other people see or the sender's per-message preview size.
Show or hide preview cards under chat messages for the caller. `false` hides both workspace entity cards and external link cards; chips and links stay in the message, and nobody else is affected.
Update the caller's per-channel notification preferences (unread_display: count | dot | since_visit).
Get a specific channel by UUID, with members.
Update a chat channel's name, topic or settings.
Delete a chat channel and its messages.
Create a poll in a channel. Creator must be a channel member.
Poll results for the poll attached to a chat message.
Poll details (question, options, the caller's votes).
Cast a vote for a poll option.
Remove the caller's vote for a poll option.
Tallied poll results (respects anonymity).
Close a poll to further voting (creator or channel admin).
Delete a poll (creator or channel admin).
List files and folders in the user's Drive. Returns items with id, name, mime_type, size_bytes, is_folder, parent_folder_id, owner info, and timestamps. Use this when the user asks to see their files, browse a folder, find a specific file, or check what's in Drive. If the user is currently viewing Drive, use context for the current folder. A `parent_id` the caller cannot view answers 404, exactly as an absent folder does; a folder shared from another workspace is browsed through `GET /api/drive/{id}/children` instead.
List the direct children of one folder, named in the path. Use this to browse into a folder — including one shared with you from another workspace, which the `parent_id` form of the list endpoint cannot return. Answers 404 for a folder the caller cannot view, exactly as an absent one does.
List only folders (not files) in Drive — useful for folder pickers. Pass root to scope the listing to My Drive, shared drives, or Shared with me.
Create a new folder in Drive. Returns the created folder with id, name, and path. Use this when the user asks to create, make, or add a new folder for organizing files. Supports nested folders via parent_folder_id.
Breadcrumb chain (root → leaf) for a folder. Powers breadcrumb hydration on deep-link reloads. Viewer permission is sufficient.
List the caller's file requests for one owned Drive folder.
Revoke a folder's public file request immediately.
Initiate a file upload. Use upload_url for direct GCS uploads. If GCS is unreachable, non-browser clients PUT raw bytes to api_upload_url on the Ambiguous API host within api_upload_max_bytes. Then confirm the returned file id. The API grant permits one admitted attempt; retry 429 after Retry-After. If a PUT result is uncertain, confirm the original file before initializing again. Browsers stay direct; do not mix the URLs.
Add a file to Drive from JSON, for clients that cannot send multipart: give either an https `url` to fetch (up to 100 MB) or the bytes as `content_base64` (up to 25 MB decoded), plus a `name`. The file goes through the same checks as a browser upload: its type is decided from the bytes, blocked types are refused, and it counts against the caller's storage quota. Links to private network addresses are refused, including through redirects. Returns the stored file.
Confirm a presigned upload completed successfully. Verifies the blob exists, rechecks its real size and stored bytes, then marks the file as confirmed and emits the upload event. A retryable storage-inspection failure leaves the upload pending and returns 503; a missing object removes the pending row and returns 410. Idempotent — confirming an already-confirmed file returns it unchanged without repeating the upload event.
Get Drive storage usage.
Storage breakdown by file type.
Recently used files for the current user — files they opened in the last 30 days (their own views, including files shared with them that they can still open) OR files they own created in the last 24 hours. Sorted newest first by their own last view, or creation for their own new files.
Files the current user is likely to want next, each with the reason it is suggested: they opened it (`opened`), someone shared it with them (`shared`), or someone else added it to a folder they own or an agent they manage created it (`added`). Covers the last 14 days, newest reason first, at most one entry per file and only files the caller can still open. Stars never influence it. Use this for the My Drive home, not for a full history; use `drive_files_recent` for everything the user opened.
Files the current user has starred.
Files shared with the current user through a direct, team or external grant, newest share first (`sort=modified` orders by last modified instead). Each row carries `shared_at` and `shared_by`. Excludes files the user owns.
Current user's tag catalog. Each tag carries name, color, and sort order.
Create a new tag in the user's catalog.
Rename, recolor, or reorder a tag.
Delete a tag (cascades to all file assignments).
List everything you tagged with the given tag: Drive files, docs/sheets/slides and forms you own, newest first. Docs and forms carry `document_type`.
Batch-fetch up to 50 files by ID. Powers drive:// embed hydration in Chat/Docs/Wiki/CRM.
Batch-fetch permissions for up to 50 files by ID. Returns a map of file_id → permissions array.
Move multiple files to a folder (max 100). The destination needs editor access (contributor in a shared drive); a folder the caller cannot see answers 404 like one that does not exist, and a visible one they cannot write to answers 403. Per-file failures come back in `errors`: "File not found" for a file the caller cannot view, "Forbidden" for one they can view but not edit. An item whose sharing is restricted (editors can't share, on it or a folder above it) moves only with its owner.
Copy multiple files to a folder (max 100). The destination needs editor access (contributor in a shared drive); a folder the caller cannot see answers 404 like one that does not exist, and a visible one they cannot write to answers 403. A copy into a shared-drive folder belongs to that drive.
Find deleted Drive file rows and folders the caller owns, trashed themselves, or holds the owner rung on (a folder above it is theirs, or they manage its shared drive). Returns paginated file IDs and metadata for restore_file or permanent_delete_file. Can include materialized native-document shadow rows, but does not list every native Docs, Sheets, Slides or Canvas item: use documents_trash_list for the complete native-document trash owned by the caller. An empty result does not mean the whole workspace trash is empty.
List shared drives user is a member of.
Create a shared drive. Creator becomes a manager.
Get shared drive details.
Update a shared drive (manager-only).
Delete a shared drive (destructive, manager-only).
List shared drive members.
Add a member (manager-only).
Update a member role (manager-only).
Remove a shared drive member (manager-only; cannot remove last manager).
List files in a shared drive.
Create a folder inside a shared drive. Requires contributor+ on the drive. Pass `parent_folder_id` to nest.
Attach one of your tags to a Drive file, a doc/sheet/slide, or a form, so one tag can group a PDF with the document it belongs to. Owner-only (anyone else gets 404) and idempotent. Pass `resource_type` to name the kind, or omit it to infer it from the id.
Detach a tag from a Drive file, a doc/sheet/slide, or a form. Owner-only. Pass `?resource_type=` to name the kind, or omit it to infer it from the id.
A file's bytes. A native item without stored bytes returns 422 native_document_not_copyable; attach it as a link. A browser is redirected to a short-lived signed storage URL where signing is available and is otherwise sent the bytes; an MCP tool call always gets the bytes, refused 413 above the MCP result size limit. Pass `?download=true` for `Content-Disposition: attachment`; default is inline. Content retrieval failures can return 503/504 via raw Response because those statuses are outside the typed contract error-code enum.
Revoke a folder gallery: restore the folder and its direct-child files to restricted and clear link settings. Owner-only. Legal-record children are left untouched.
Restore a file from trash. Needs the owner rung (the owner, the owner of a folder above it, or a manager of its shared drive); a caller who can see the file without it gets 403. An item whose permanent deletion has started, or a folder holding one, cannot be restored: 409 FILE_PURGE_IN_PROGRESS.
Export a workspace document (Doc, Sheet, or Slide) to a file format via its Drive ID. Docs support: docx, pdf, html, md. Sheets support: xlsx, csv. Slides support: pptx, pdf, html. Binary Drive files (images, archives, etc.) cannot be exported — use download instead.
Get detailed information about a file or folder in Drive. Returns id, name, mime_type, size_bytes, storage_key, owner info, parent folder, and timestamps. Use this when the user asks about a specific file's details, size, type, or location. If the user is currently viewing a file in Drive, use the resource ID from page context.
Rename, move, or star/unstar a file. Rename and move require edit access; stars are owner-only. A move also needs editor access on the destination folder (contributor in a shared drive), and an item whose sharing is restricted moves only with its owner (403 `drive_move_requires_owner`).
Delete a file or folder from Drive. This is a destructive action — always confirm with the user before executing. Deleting a folder also deletes all files inside it. Use this when the user explicitly asks to delete, remove, or trash a file or folder.
Start connected-mailbox (Gmail) OAuth — returns the consent URL.
List connected mailboxes.
Disconnect a connected mailbox.
Verify a connected mailbox by fetching its Gmail profile.
Full-text + operator search (Gmail-style: from:, to:, cc:, bcc:, subject:, has:attachment, is:read|unread|starred, in:inbox|sent|drafts|trash|archive, label:, after:YYYY-MM-DD, before:YYYY-MM-DD, size:N, larger:N, smaller:N, "phrase"). cc:/bcc: filter by recipient type. size:/larger:/smaller: filter by body byte size.
Suggest a short continuation of an email draft being written. Takes the text typed so far, optionally the subject and the earlier messages of the thread being replied to, and streams the suggested next words.
Export a message as RFC 5322 EML, returning retained source octets when available.
Per-folder badge counts for sidebar Returns { inbox: unread_count, drafts, scheduled, snoozed } — Gmail-style.
Unread mail summary for a cross-module unread view: the total unread count plus the top N (default 5, max 25) unread inbox rows with sender, subject, preview and timestamp, so a unified list can render without pulling the full inbox.
List available send-as addresses for current user
Set the user's default send-as address.
List all aliases for current user's mailbox
Add a new alias on a verified workspace domain
Set alias as default send-as identity
Delete a send-from alias address by id.
List email submissions awaiting a human decision. Returns the caller's own submissions plus the ones they may decide — as the submitting agent's current manager, or as a workspace admin/owner. A former manager sees nothing.
Read one submitted email exactly as it will send — From, Reply-To, the complete To/Cc/Bcc set, subject, body with signature applied, attachment manifest, threading and requested schedule. A caller without access gets 404, never 403.
Presigned URL for one frozen attachment, addressed by its position in the manifest. The bytes are the copy taken at submit time, so a later edit of the source Drive file does not change what the reviewer reads. The URL is minted only after the review access check.
Approve and send, request changes, reject, or (as the sender) withdraw one submitted email. The decision names the revision it applies to: exactly one decision per revision succeeds, and a decision on a superseded revision conflicts rather than releasing content the caller has not read. Approving records the human and time against that revision; it authorizes nothing else — not the thread, not a reply, not a later email.
List the shared mailboxes the caller holds, or every one in the workspace with `scope=managed` (admin only)
Create a shared mailbox and address it on a workspace domain (admin only)
Get one shared mailbox and its members
Rename a shared mailbox's display name; its address is unchanged (admin only)
Retire a shared mailbox (admin only). Soft: stored mail is retained and the address stays reserved.
Restore a retired shared mailbox (admin only). Total — delete cascaded nothing.
The people who can hold a thread in this shared mailbox — team grants flattened to their members.
Grant a user or a team access to a shared mailbox (admin only)
Revoke a user's or team's access to a shared mailbox (admin only)
Get the effective email signature.
Set personal email signature. Pass null to clear.
Finalize an unchanged saved draft with saved_draft: {id, revision} from its authorized detail Read; this reuses that email and starts no second Undo window. A stale or consumed occurrence returns 409; reconcile the same email instead of creating another send. Otherwise, send email as the current user: Markdown becomes HTML; To/Cc/Bcc use email addresses. For workspace-only agents, any external To/Cc/Bcc queues the WHOLE email for a human decision; no recipient gets delivery. Response delivery_status:pending_approval with pending_approval.review_id. Report awaiting approval, never sent; do not resend. A native Drive document attached without stored bytes returns 422 native_document_not_copyable; attach it as a link. A suppressed recipient returns 400 recipient_suppressed; replaying an older suppressed send without a saved reason returns 400 delivery_suppressed. The workspace external-send cap returns 429 external_send_limit_exceeded.
List the user's received emails (inbox). Returns emails with id, from, subject, preview, read status, and sent timestamp. Use this when the user asks to check their inbox, see recent emails, or find a specific email. Use detail='headers' for bandwidth-efficient scanning (omits body), or 'minimal' for id + labels only. Use label_ids to filter by specific labels (AND semantics — all listed labels must be present).
List the emails sent from this mailbox.
List the starred emails in this mailbox.
List the snoozed emails in this mailbox.
List archived threads (one per thread)
List muted threads (inbox view for muted conversations)
List trashed (soft-deleted) emails — the caller's own rows in folder='trash'. Restore one with `untrash_email`; nothing is purged automatically.
List the unsent draft emails in this mailbox.
Create a draft email without sending it. The draft is saved in the drafts folder for later review, editing, or sending. Use this when the user wants to compose an email for later, save a draft, or prepare a message without sending it immediately. A native Drive document attached without stored bytes returns 422 native_document_not_copyable; attach it as a link.
Update an existing draft email. Only works on drafts (not sent messages). Use this to refine recipients, subject, or body of a draft across multiple iterations before sending. A native Drive document attached without stored bytes returns 422 native_document_not_copyable; attach it as a link.
Use to send a saved draft. Normal delivery moves it from drafts to sent, triggers SES and creates recipient inbox copies. For a workspace-only agent, any external To/Cc/Bcc instead queues the WHOLE email for human decision; nobody receives it. Response delivery_status: pending_approval and pending_approval.review_id. Report awaiting approval, not sent; never resend. A native Drive document attached without stored bytes returns 422 native_document_not_copyable; attach it as a link. A suppressed recipient returns 400 recipient_suppressed. The workspace external-send cap returns 429 external_send_limit_exceeded.
List scheduled (unsent) emails
Change a pending delivery time, or remove its schedule with scheduled_at:null and expected_scheduled_at to retain the message as a draft.
Get undo-send duration preference.
Set undo-send duration (seconds).
Undo an in-flight send within the undo-send window.
Get read-receipts default for new compose.
Toggle the default for new compose windows
Get the user's custom snooze quick-presets. Returns defaults when no custom presets are saved.
Set the user's custom snooze quick-presets (up to 10). Pass an empty array to reset to defaults.
Get the caller's own mail-row swipe actions. Returns the defaults (right archive, left trash) when nothing is saved.
Set the caller's own mail-row swipe actions — one action per direction, chosen from archive, trash, read, star, snooze.
Fetch open events for an email (sender only) Returns the rollup (first/last seen, open count) plus up to the most recent 50 individual open events. 403 if the caller doesn't own the email's mailbox.
List emails with a specific label
Read your rule-management receipt, Undo deadline and reported conflicts.
Restore uncontested rule configuration once within the server Undo window. Preserve later edits and labels now used elsewhere; report conflicts. Messages are never changed.
Read progress, the first page and count of field conflicts, and the server undo_until for your own rule application.
Page the message fields a restoration left alone because they changed after the rule applied.
Stop an active application and restore its uncontested changes in the background; the rule remains enabled.
Stop application and restore uncontested prior fields exactly once; expired Undo is refused visibly.
Resume a failed transactional job from its durable progress; committed messages are never reapplied.
Map supported search terms to conditions and name every omitted restriction. Preview changes nothing.
Preview size-limited Gmail API filter JSON plus a label map or an Ambiguous backup. XML must be normalized by the browser; unsupported restrictions skip the whole filter.
Return serialized file bytes as text plus every Gmail loss and omission; Backup preserves writable configuration, on/off and order.
Download mailFilters.xml or an exact Ambiguous JSON backup. Use preview_mail_rule_export first to inspect every Gmail loss and omission.
List mail rules / filters.
Preview the caller's received Inbox/Archive messages with the delivery matcher. Counts and pages matches within the newest candidates, showing the five newest by default. more_may_match means older messages were not read; applying the rule still covers them.
Opt in to a durable background application of this rule to every received Inbox/Archive message received at or before submission that matches it, with no cohort cap; mail received later is handled by the saved rule itself. The job starts as collecting, with total growing until collection ends, then applies; read progress with get_mail_rule_application. Forward is unavailable, Trash needs confirm_trash, and notification-only rules cannot apply. Retry the same request with the same idempotency_key. Explicit retire_unaccepted:true reconciles that exact original request: a matching existing job is returned unchanged; a different request returns 409. If no job exists, retirement durably spends the caller's personal-mailbox key as an empty cancelled job with no rule reference. Retirement does not read or authorize the requested rule, does not cancel admitted work, and excludes the retirement flag from request identity.
Reorder the mail rules to change which runs first.
Get one mail rule's conditions and actions by id.
Delete one mail rule by id.
List the saved mail templates in this mailbox.
Create a reusable mail template for composing.
Get one saved mail template by id.
Update a saved mail template's subject or body.
Delete one saved mail template by id.
Record one use of a saved mail template by incrementing its usage count and updating its last-used timestamp. Returns the template; does not compose or send an email.
List every mail label in this mailbox.
Per-label unread counts for sidebar badges. A separate endpoint so the sidebar can refresh counts on read/receive without re-fetching the label definitions themselves. Shape mirrors /folder-counts.
Create a mail label for organizing threads.
Reorder the mail labels shown in the sidebar.
Update a mail label's name, color or sidebar order.
Delete one mail label from this mailbox.
Get vacation auto-reply settings.
Update vacation auto-reply settings.
List all blocked senders for current user's mailbox
Block a sender (address or domain).
Unblock by raw pattern (convenience for "Unblock sender" button)
Unblock by id or pattern.
Unblock a sender by their blocklist entry id.
One-click unsubscribe. Triggers the List-Unsubscribe URL (RFC 8058) on the user's behalf. Falls back to telling the client to open a mailto: link if no HTTP URL is present.
List forwarding addresses + recent log.
Remove a forwarding address
Get all emails in a thread and its muted flag. By default, marks the thread read and updates unread counts; pass mark_read=false for a passive read.
Check if a thread is muted
Archive every inbox-folder message in a thread
Trash all messages in a thread atomically. Moves every message in the thread to the trash folder.
Restore all trashed messages in a thread to inbox. Only affects messages currently in trash.
Archive multiple emails in one atomic call. Moves all specified messages to the archive folder. Use this when the user wants to bulk-archive many emails (e.g., inbox-zero cleanup, archiving all messages from a sender).
Trash multiple emails in one atomic call. Moves all specified messages to the trash folder. Use this when the user wants to bulk-trash many emails (e.g., spam cleanup, clearing a project mailbox).
Batch modify multiple emails in one atomic call. Can star/unstar, mark read/unread, add/remove labels, and change folder — all at once for up to 500 messages. Use this when the user wants to bulk-update many emails (e.g., star all from a sender, mark a batch as read, apply labels to multiple messages).
Apply mark_read, mark_unread, star, unstar, archive, unarchive, trash, untrash, label_add or label_remove to up to 500 emails selected by IDs or search_query. Label actions take payload.label. threaded=true includes whole conversations.
Get calendar invite embedded in an email.
RSVP to a calendar invite on an email.
List attachments for an email
Get an authenticated attachment content URL. Access is checked again when the content is read.
Restore a trashed email to inbox
Snooze an email until a datetime.
Cancel a scheduled email
Get a single email by ID, including full Markdown body, sender, recipients (to/cc/bcc), subject, and timestamps. Use this when the user wants to read an email, see its contents, or reply to a specific message. If the user is currently viewing an email, use the email ID from the page context. Use detail='headers' for envelope fields only (no body), or 'minimal' for id + labels only. By default, marks the email read and updates unread counts; pass mark_read=false for a passive read.
Update email status (read/unread, starred, snoozed)
Delete one email message by id.
Add an image to task description. Returns the parent-owned image URL `/api/assets/{id}/content`. Input: content_base64 bytes, source_file_id (copies a viewable Drive file), or source_url (copies public HTTPS through SSRF guard); no source link retained. Native sources without stored bytes return 422 native_document_not_copyable; attach it as a link. Byte-sniffed images only, 415 for other types or malformed SVG. SVG sanitized on upload and every read, inline image/svg+xml with restrictive CSP; malformed stored SVG returns 404. Drive SVG stays attachment. Owner quota exceeded 413; same image deduplicated per parent. Absent from Drive, readable exactly with parent.
List task templates in the workspace. Templates pre-fill fields when creating tasks. Optionally filter by project. Returns name, default fields (priority, status, estimate, assignee, labels), and creator.
Create a new task by applying a template's defaults. The template pre-fills status, priority, estimate, assignee, description, and labels. Override any field by passing it explicitly. Title is always required.
Get a single task template by ID.
Update a task template. Workspace admins or project admins only.
Delete a task template. Workspace admins or project admins only.
Export authorized tasks matching the selected filters as CSV (default) or JSON. CSV includes human-readable columns and a task_json column containing the complete public task representation. Owned description images are embedded; an unavailable or malformed owned image refuses the export with its task reference.
Batch-fetch up to 50 tasks by ID. Powers LinkedTaskItem hydration in Docs/Wiki.
List task labels visible in the workspace, grouped by `group_name`. Without `project_id`, returns workspace-level labels only. With `project_id`, returns workspace-level labels plus labels scoped to that project.
Create a workspace task label. Returns 409 if a label with the same name already exists in the group.
Update a task label.
Delete a task label. Existing assignments are removed.
Get the workspace's SLA thresholds (response + resolution times per priority). Auth-only.
Aggregate SLA metrics for the workspace (compliance rates, avg times, counts per status). Optionally scope to a project.
List task views accessible to the caller (own + shared + system defaults).
Create a saved task view (filter + layout). Setting `is_shared: true` exposes it to the workspace.
Update a saved task view (rename, change filters or layout defaults, toggle sharing).
Delete a saved task view.
Get the caller's layout preferences for a view, or fall back to the view's defaults. `source` indicates which.
Upsert the caller's layout preferences for a view (per-project scope optional).
List task IDs the caller is subscribed to (manual + auto-subscribed via creator/assignee/comment).
Bulk-update up to 100 tasks. Tasks you can't edit are skipped; `updated` counts the rest. Omitted fields stay unchanged; pass `null` to clear nullable fields.
Soft-delete multiple tasks at once by ID. Only deletes tasks the current user created. Max 100 task IDs per call. Subtasks cascade. Use `restore_task` to recover individually.
Archive up to 100 tasks at once. Only archives tasks the caller created or is assigned to.
Count of triage-bucket tasks (untriaged + projects in 'triage' status). Powers the sidebar inbox badge.
List your Tasks-sidebar pins: saved views and projects, not starred documents, Drive items, channels or teams.
Pin a saved task view or project in your Tasks sidebar. For document, Drive, channel or team favorites, use set_resource_starred.
Unpin a saved task view or project from your Tasks sidebar, matching resource_type and resource_id. For other resource favorites, use set_resource_starred.
Reorder your Tasks-sidebar view/project pins using ordered_ids (favorite-row IDs, top-to-bottom), not resource IDs.
Resolve a workspace-scoped task key (e.g. `TASK-123`) to its task ID. Used by deep links and integrations.
Batch-fetch every relation type for up to 100 tasks. Powers relation overlays and sweeps.
List tasks in the workspace. Pass assignee_id to filter by assignee (use the caller's user ID for 'my tasks'). Supports filtering by status, priority, project, and free-text query. Defaults retain full content and aggregates. Use fields= for sparse rows and include_counts=false on cursor continuation pages to avoid repeated aggregate work.
Create a new task. Auto-subscribes the creator and (if assigned) the assignee. Auto-sets SLA due date when priority is set without an explicit due_date. Auto-logs a CRM activity if linked to a contact.
Assign a label to a task. Returns the task's full label list.
Remove a label assignment from a task. Returns the task's remaining label list.
List all users subscribed to a task (with the reason each is subscribed).
Remove another user's subscription from a task. Requires task access.
Check whether the current user is subscribed to a task (and the reason if so).
List the collaborators on a task — the people holding `editor` on it (and its subtasks) without a role on its project.
Share a task with one workspace member: grants `editor` on this task and its subtasks (never on its project) and subscribes them. A projectless task is shared by its creator or assignee; a project task by an editor on that project.
Revoke a task collaborator's grant. The add-time subscription is left in place — it grants no access on its own.
List the pull requests linked to a task that you can see: a link shows only when you can read both the task and the repository.
List immediate subtasks of a task (`parent_task_id = :id`).
Create a subtask under the given parent task. Equivalent to `POST /api/tasks` with `parent_task_id` pre-filled from the URL parameter.
Start a time-tracking timer on a task for the current user. If another timer is already running for this user, it is stopped automatically (only one active timer per user).
Stop the current user's active timer on this task and return the stopped entry. Adds its duration to time_spent_minutes. Returns 404 if this user has no active timer on the task.
List audit-log activity entries for a task (status changes, assignments, comments, label edits, etc.). Paginated via `limit`/`offset`.
List edges in the task-relations graph for a task (related, blocking, blocked_by, duplicate).
Create a relation between two tasks. `blocking`/`blocked_by` auto-creates the inverse edge. With `type=duplicate, mark_as_duplicate=true` the source task is cancelled with a system comment.
Delete a task relation edge. Auto-deletes the inverse edge for blocking/blocked_by pairs.
Restore a soft-deleted task when the caller has task editor access.
Clone a task — copies title (appends ' (copy)'), description, priority, project, goal, cycle, estimate, due/start date, recurrence, custom fields, and labels. The clone starts at status 'todo' with a fresh task key.
List task-owned attachments and existing legacy attachments.
Copy a Drive file you can view into a task-owned attachment. The original remains independent. Re-attaching the same source returns the active attachment; a stable idempotency_key prevents replay after removal. A native doc, sheet, deck, canvas or form without stored bytes returns 422 native_document_not_copyable; attach it as a link.
Remove a task attachment and its owned bytes. Removing a legacy attachment leaves its original Drive file unchanged.
Get a single task by ID, including full details: title, description, status, priority, assignee, due date, labels, and subtask count.
Read the caller's task editing, project-move and sharing permissions, with editable IDs for the first 200 immediate subtasks.
Update a task. Status changes to 'done' auto-spawn the next recurrence (when set), trigger CRM activity logging, and emit task.completed events.
Soft-delete a task. Only the creator can delete.
List project templates in the workspace.
Create a project template.
Create a new project from a template. Pass `{name, description}` in the body.
Get a single project template.
Delete a project template.
List accessible projects in the current workspace, optionally filtering by a case-insensitive name query before pagination. Returns each project with name, icon, color, visibility, and task count. Use this when the user asks to see their projects or find a project by name.
Create a new project to group related tasks. A bound document or Wiki page atomically selects the new project; named source editor access is required. An already-bound source or a bound project with workspace visibility is refused.
Get a single project by ID, including task count, visibility, and metadata.
Update a project's name, description, icon, or color. Provide only the fields you want to change.
Delete a project by ID. Only project admins or workspace admins can delete. Tasks belonging to the project are not deleted — they become unassigned from the project.
List project members.
Add a member. Body `{user_id, role?}` — or `{team_id, role?}` for a team share. A project editor may add, capped at its own role: granting a role above it is refused `403`. Notifies the new member.
Change a member's role.
Remove a grant from the project. The path id is the principal id: a user by default, or a team with `?type=team`. Notifies the removed user (unless self-removal). The Everyone team's row is refused — workspace access is set through the project's visibility, which is its only control.
List status updates for a project, newest first. Returns the health timeline with author and timestamp.
Post a status update on a project. Use this when the user wants to report progress, flag a risk, or share a status update on a project. The status field captures the health signal (on_track, at_risk, off_track, paused, completed, cancelled).
Edit a status update. Author or admin only.
Delete a status update. Author or admin only.
Get the caller's notification preference for the project.
Set the caller's notification preference (`level`) for the project.
List cycles in the active workspace. `?status=` filters by status; `?limit=` (1–200) and `?offset=` paginate. `has_more` is set when the underlying query returned a hint of more rows.
Create a new cycle. Body fields: name, start_date, end_date, status?, cooldown_days?, auto_rollover?.
Cross-cycle velocity (last N closed cycles). Default `limit=10`.
Get one task cycle and its dates by id.
Update cycle fields.
Delete a cycle. Tasks attached to the cycle are unlinked, not deleted.
Burndown + summary analytics for the cycle.
Complete a cycle. If `auto_rollover` is set, incomplete tasks are moved to the next cycle.
Add tasks to the cycle. Body `{task_ids: [...]}`.
Remove tasks from the cycle. Body `{task_ids: [...]}`.
List editable table definitions by worksheet.
Create or edit one table. Bounds describe DATA only; totals follow data. Current headers win; cell exceptions remain.
Change supplied table fields only. Identity and unrelated declarations remain.
Remove a table definition while retaining its cells.
List workbook custom table styles.
Create or edit a named custom table style from differential elements. False and absent settings differ. expected_updated_at refuses stale replacements with 409; omission deliberately replaces.
Delete an unused custom table style.
Set all twelve workbook theme colors; affects every theme-linked workbook feature. Null restores native/default palette. expected_updated_at refuses stale replacements with 409; omission deliberately replaces.
List editable sparklines and their owning worksheet.
Create a native editable sparkline in one cell from a worksheet row or column.
Edit only the supplied sparkline fields while preserving unrelated workbook edits.
Delete one editable sparkline without changing its source cells.
List the current user's spreadsheets (owned + shared). Convenience alias for `GET /api/documents?type=sheet`. Returns the lightweight `SheetSummary` shape without the structured `data` body — call `GET /api/sheets/:id` to load a specific sheet's tabs/columns/rows.
Create a new spreadsheet. Returns the full `Sheet` envelope (id, title, owner, visibility, timestamps, and `data` body) so the caller can immediately read or edit cells without a second round-trip. Optional `content` may be a SheetData JSON object or a string that parses as JSON.
Append rows of values to a sheet at the first empty row in a range. The server finds the insertion point — no need to read the last row first. Uses OVERWRITE semantics (does not shift existing data).
Fetch multiple disjoint ranges from a spreadsheet in a single request. Avoids N sequential round-trips when reading KPIs from different regions.
Clear all cell values in multiple ranges atomically. Sets cells to null without removing rows or columns. Useful for resetting template sheets or clearing data regions.
Copy a sheet tab from one spreadsheet to another. Deep-copies columns, rows, formulas, formatting, and merges. Cross-document formulas will show as #REF!. Requires viewer access on source and editor access on destination.
Create a named range or named expression on a spreadsheet. Send either an A1 range or an exact literal Excel expression, such as a structured table reference. External-workbook expressions are preserved verbatim and remain unresolved.
List named ranges and named expressions on a spreadsheet. Each entry returns its name and either an A1 range or an exact literal expression. External-workbook expressions remain unresolved.
Delete a named range from a spreadsheet by name. Names are matched case-insensitively.
Update a named range or named expression with either an A1 range or an exact literal Excel expression; optionally rename it with newName. Names are matched case-insensitively. External-workbook expressions are preserved verbatim and remain unresolved.
Set a data-validation rule on a spreadsheet cell range in A1 notation. Supports dropdown/range lists, numeric/text/date conditions, checkbox values, and custom formulas. Numeric constraints use JSON numbers, not numeric strings. Strict configures invalid-entry rejection rather than warnings in the editor; API cell writes do not enforce these rules.
Remove a data validation rule from a spreadsheet by range. The range is matched after normalization (sheet tab defaults to Sheet1 if not specified). Returns 404 if no rule exists for that range.
List all data validation rules on a spreadsheet. Returns each rule's range (A1 notation) and rule definition (type, values, strict, inputMessage).
List all conditional formatting rules on a spreadsheet. Returns each rule's ID, ranges (A1 notation), and one canonical rule payload.
Add a conditional formatting rule to a spreadsheet. Provide one of booleanRule, gradientRule, iconSetRule, or dataBarRule. The server generates a unique ID for the rule.
Update a conditional formatting rule by ID. Partial update: ranges or one canonical rule payload; replacing the kind clears the previous kind.
Delete a conditional formatting rule from a spreadsheet by its ID. Returns 200 on success, 404 if the rule does not exist.
List all protected ranges on a spreadsheet. Returns each range with its ID, A1 range, description, warningOnly flag, and editorUserIds allowlist.
Protect a range in a spreadsheet. Protected ranges restrict editing to users in editorUserIds. Set warningOnly to true for soft warnings instead of hard blocks.
Remove a protected range from a spreadsheet by its ID. Returns 200 on success, 404 if the range does not exist.
Get a spreadsheet's structured data including columns, rows, and formulas. Returns the `Sheet` envelope: top-level metadata (id, title, owner, visibility, timestamps) plus the structured sheet body nested under `data` (see `SheetData`). Read the tab list as `response.data.sheets`; columns and rows are under `response.data.sheets[0]`, not at the response root. If the user is currently viewing a sheet, use the sheet ID from the page context.
Get the spreadsheet as the structured JSON canvas. Returns the `Sheet` envelope: top-level metadata plus the structured sheet body nested under `data` (see `SheetData`). Read the tab list as `response.data.sheets`; columns and rows are under `response.data.sheets[0]`, not at the response root.
Use to read a rectangular slice, max 10,000 cells. Returns 2D values and resolved start/end letters+rows; orientation=COLUMNS yields column-major inner arrays. valueRenderOption: STORED(default) preserves raw values/formula strings; FORMULA forces formulas; EVALUATED runs server engine (SUM(A1:A3)→6). Cross-tab Sheet2!A1 resolves; cross-document IMPORTRANGE returns#REF!.
Update specific cells in a spreadsheet. Provide an array of cell updates, each specifying the row index (0-based), column ID, and new value. Optional expected_content_sha256 checks the server-issued revision of the entire workbook; preconditions check stored cells and row_count_preconditions check exact tab row counts. All supplied checks run under the mutation lock. A mismatch, or a tab or column the sheet no longer has (TAB_NOT_FOUND, COLUMN_NOT_FOUND), returns 409 without changing values or formatting. If the user is currently viewing a sheet, use the sheet ID from the page context.
Replace the complete merged-cell range collection for a sheet tab using zero-based inclusive startRow, startCol, endRow and endCol coordinates. Read the current sheet and retain unrelated entries when changing one merge. An empty merges array clears all merges on that tab.
Add a new tab to a sheet. Auto-generates the name if not provided.
Rename an existing sheet tab.
Remove a sheet tab by name. Cannot remove the last remaining tab.
Replace a sheet tab's columns and/or rows wholesale. Use for bulk data replacement. Use update_cells for surgical edits.
Set worksheet validation prompt suppression without replacing cells, rules, or presentation. Send disablePrompts=null to restore the absent default setting.
Set a column's declared type without coercing cells. expected_type must match its current type under the document lock; null denotes absent type metadata. Conflicts return 409 without changing the workbook.
Add a column to a sheet tab.
Remove a column from a sheet tab by column ID. Use `?sheet=` to target a tab other than the first.
Delete columns atomically and return the bounded change receipt for undo or redo.
Append rows to a sheet tab. Each row is an object keyed by column ID. Example: [{"A": "Widget", "B": 9.99}]. The response carries `history`, a receipt for `restore-structure` undo and redo.
Delete rows from a sheet tab by 0-based index.
Sort a tab's rows by one column, optionally within an exact filter range. Re-anchors formulas, comment anchors, and cell notes so nothing orphans.
Move one or more consecutive columns to a new position. The response carries a history receipt for Undo. Rewrites dependent formulas so they follow the moved column to the same logical data, and re-anchors notes and comment anchors.
Move one or more consecutive rows to a new zero-based position, carrying its values, formatting, geometry and dependent references.
Undo or redo a structural change receipt without overwriting unrelated later edits.
List data connectors registered on a sheet. Connectors fetch data from external sources (URL, CSV, etc.) into specific cells.
Get a single data connector by ID, including its config and last-run status.
Delete a data connector. Does not affect the cells it previously wrote to.
List recent execution history for a data connector. Each row records one manual or scheduled run with its status, duration, row count, and error (if any). Use the `limit` query param to cap the page size (default 20, max 100). Results are ordered by completion time, newest first.
List every pivot table across every tab of the sheet. Each entry is annotated with the `tab` name it lives in.
Create a pivot table from sheet data. A new sheet tab is automatically created to hold the pivot. Specify the source data range, which columns to group by (rows/columns), and which columns to aggregate (values).
Compute the result of a persisted pivot table — applies row/column grouping, value aggregation, sort orders, and any matching slicer filters, and returns the rendered cells. The pivot config itself (row/column/value fields, source range) is not modified.
Patch a pivot table's config, including native presentation and preferred cache identity. Only the fields you send are updated; the rest of the pivot config is preserved. Send `nativeCacheId: null` to clear its authored preference. Pass a `tab` field to move the pivot to a different tab.
Remove a pivot table from the sheet. The host tab itself stays — only the pivot row in `pivotTables[]` is removed. Returns `{ deleted: true }` on success.
List editable picture groups with cell anchors and CSS pixel geometry. Every leaf references an image asset owned by this sheet; native group transforms use EMU coordinates.
Create editable picture groups with cell anchors and CSS pixel geometry. Every leaf references an image asset owned by this sheet; native group transforms use EMU coordinates.
Update editable picture groups with cell anchors and CSS pixel geometry. Every leaf references an image asset owned by this sheet; native group transforms use EMU coordinates.
Delete editable picture groups with cell anchors and CSS pixel geometry. Every leaf references an image asset owned by this sheet; native group transforms use EMU coordinates.
List editable floating images with cell anchors and CSS pixel geometry. Image src must reference an image asset owned by this sheet.
Create editable floating images with cell anchors and CSS pixel geometry. Image src must reference an image asset owned by this sheet.
Update editable floating images with cell anchors and CSS pixel geometry. Image src must reference an image asset owned by this sheet.
Delete editable floating images with cell anchors and CSS pixel geometry. Image src must reference an image asset owned by this sheet.
List editable floating shapes with editable text, cell anchors and CSS pixel geometry.
Create editable floating shapes with editable text, cell anchors and CSS pixel geometry.
Update editable floating shapes with editable text, cell anchors and CSS pixel geometry.
Delete editable floating shapes with editable text, cell anchors and CSS pixel geometry.
List every chart across every tab of the sheet. Each entry is annotated with the `tab` name it lives in.
Embed a chart on a tab of the sheet. Charts visualize a rectangular data range — pick a chart type (bar / line / pie / scatter), the source range, and an overlay position. Returns the created chart with its server-assigned id.
Patch a chart's config — type, title, data range, overlay position, or series styling. Only the fields you send are updated; the rest of the chart config is preserved.
Remove a chart from the sheet. Returns `{ deleted: true }` on success.
List every slicer across every tab of the sheet. Slicers are interactive filter widgets that filter pivots / charts on a shared source range. Each entry is annotated with the `tab` name it lives in.
Add a slicer to a tab of the sheet. Slicers filter pivot tables and charts that share the same source range by the values selected for a single column. `selectedValues: null` passes everything through; `[]` filters everything out; a non-empty array passes only those values. Returns the created slicer with its server-assigned id.
Patch a slicer's config — source range, target column, selected values, overlay position, or title. Only the fields you send are updated; the rest of the slicer config is preserved.
Remove a slicer from the sheet. Returns `{ deleted: true }` on success.
List the slide-level layouts the editor knows about.
List the built-in slide deck templates.
Save an existing deck as a reusable workspace template. Deep-copies the deck with fresh element IDs into a new workspace-visible template document; the source deck is left untouched. Returns the new template's ID.
Start a new deck from a template — built-in or workspace. Deep-copies the template deck with fresh element IDs into a new deck, substitutes `{{key}}` placeholders to bracketed prompts (e.g. `{{client_name}}` → `[Client Name]`), and returns the new deck's ID.
List IDs of slide decks the current user has starred.
Create a new presentation deck. Omit `content` for a deck with one blank slide, or pass deck JSON to create it with its slides in place: `{ slides: [{ elements: [{ type: "text", x: 80, y: 60, w: 800, h: 80, text: "Q1 review", fontSize: 44 }] }] }`. A slide is 960 by 540 units with a top-left origin, and elements take the shape `add_slide` accepts. Returns the created `SlideDeck` envelope.
Get a presentation's full data including all slides and their elements. Returns the `SlideDeck` envelope: top-level metadata (id, title, slide_count, owner, visibility, timestamps) plus the structured deck body nested under `data` (see `SlideDeckData`). Read settings/slides/masterSlides as `response.data.slides[…]`, not at the response root. If the user is currently viewing a presentation, use the presentation ID from the page context.
Use to add a JSON slide with layout/background/elements. Each typed text/image/shape/table/video/group has x, y, w, h and type fields. Give EACH title/bullet/paragraph its own positioned text element, never multiple lines/bullets together. Use • bullets, not Markdown * or -. Set fontSize/fontWeight hierarchy. If viewing a presentation, use its ID from page context.
Update an existing slide in a presentation. Specify the slide by its 0-based index. You can update the slide properties (background, notes) and/or replace its elements. If the user is currently viewing a presentation, use the presentation ID from the page context.
Remove the slide at the given zero-based index.
Link a master slide to a specific slide by 0-based index. The slide stores a by-reference link, retains its existing elements as child-owned content, and clears prior master-element override markers. Current master elements inherit at render time and future master edits flow through automatically.
Group existing top-level elements on a slide into a single group. The group's bounding box is computed automatically from the children. Children positions become relative to the group origin. At least 2 element IDs required. A named element may itself be a group — `groupElements` validates only that each id is top-level on the slide, so groups nest.
Dissolve a group element on a slide back into independent top-level elements. Children positions are converted from group-relative to absolute coordinates. The group element is removed and its former children become top-level elements on the slide.
Deep-copy a slide within the same deck. The duplicate is inserted immediately after the source slide. All elements (including group children) receive new unique IDs to avoid collisions. Useful for template-based workflows: duplicate a template slide, then update the copy with new content.
Get the PNG thumbnail bytes for a specific slide. Sizes: small (200px), medium (800px, default), large (1600px). A render with an exhausted transient image read has no ETag and Cache-Control:no-store; deterministic placeholders keep their validator. A busy renderer answers 503 slide_thumbnail_capacity_exceeded with Retry-After.
Detach a slide from its master. The slide keeps its current rendered elements but no longer inherits from the master.
Apply a slide transition to every slide in the deck. Set transition.type to 'none' to clear transitions on all slides. Prefer duration_s (float seconds, 0.1–10); legacy speed presets map to slow=1.0s, medium=0.5s, fast=0.3s.
Reorder slides in a presentation. Provide the full list of slide indices in the desired new order. Example: [2, 0, 1, 3] moves the third slide to the front.
Update deck-level settings: theme, aspect ratio, default background. Changing aspect ratio rescales every slide's elements proportionally; changing default background updates slides currently using the old default.
List the master slide templates registered on a presentation. Child slides link to masters by reference and resolve current master elements at render time, beneath child-owned elements.
Create a new master slide for a deck. If no elements are provided, a default Title + Subtitle layout is used.
Update a master slide's name, background, or elements. Slides using the master resolve the updated elements by reference.
Delete an unused master slide. Returns 409 with the referencing slide count while the master is in use.
Find and replace text across all slides in a presentation. Searches every text element and table cell on every slide. Useful for template-based workflows: create a deck from a template with placeholders like {{company_name}}, then replace them programmatically. Plain string matching only (no regex).
Set the heading and/or body font across a whole presentation: every text box, shape and table on every slide and master, group children and per-run rich-text fonts included. Text at 28px or larger takes the heading font; smaller text and tables take the body font. Fonts are a closed list. The saved deck is re-read before responding: a 409 means some elements still use another font, so the change is incomplete and must not be reported as done.
Replace text placeholder elements with an image across all slides. Finds text elements whose text matches the search string and replaces them with image elements at the same position and size. Matching placeholder shapes such as {{logo}} are removed and replaced with images.
Get the PNG thumbnail bytes for the first slide (cover image) of a presentation. Convenience endpoint — equivalent to get_slide_thumbnail with index 0. A render with an exhausted transient image read has no ETag and Cache-Control:no-store; deterministic placeholders keep their validator. A busy renderer answers 503 slide_thumbnail_capacity_exceeded with Retry-After.
Apply a slide template — built-in or workspace — to the deck or to a single slide.
Create a new empty canvas for an infinite whiteboard. Use when starting a fresh flowchart, product map, or swimlane flow; returns the canvas envelope with an empty `data.elements`, then add objects with create_canvas_element. Coordinates: board units, top-left origin, +x right / +y down. Example: create_canvas({ title: "Q1 product map" }) → { id: "<boardId>", data: { elements: [] } }.
Get a full canvas by id, including every element in `data.elements`. Use when you need the complete board; for a large board prefer get_canvas_manifest (ids/types/bounds only) or get_canvas_frame to stay within context budget. Coordinates: board units, top-left origin, +x right / +y down. Example: get_canvas({ id: "<boardId>" }) → the board with all elements.
Update canvas-level metadata (title, preview cover). Use to rename a canvas or set its preview; never touches elements — use the element ops for board content. Coordinates: board units, top-left origin, +x right / +y down. Example: update_canvas({ id: "<boardId>", title: "Renamed board" }).
Control a board's shared brainstorm countdown. Use to time-box a round: start (with duration_ms), pause (banks remaining time), resume (from banked time), or reset (clear). The server stamps `ends_at` from its own clock, so every participant renders the same countdown regardless of local clocks; the timer touches no elements. Coordinates: board units, top-left origin, +x right / +y down (unaffected by the timer). Example: update_canvas_timer({ id: "<boardId>", action: "start", duration_ms: 300000 }) starts a 5-minute countdown.
Get a bounded manifest of a board — element ids, types, and bounds only (no style/content). Use this to survey a large board cheaply, then fetch a specific frame or the full board for detail. Coordinates: board units, top-left origin, +x right / +y down. Example: get_canvas_manifest({ id: "<boardId>" }) → { element_count, elements: [{ id, type, x, y, w, h }] }.
Get a frame (or swimlane) element and its children — every element whose `frameId` points at it. Use to read one region of a large board in full detail. Coordinates: board units, top-left origin, +x right / +y down. Example: get_canvas_frame({ id: "<boardId>", frameId: "<frameId>" }) → { frame, elements: [...] }.
Use to refresh changed elements: ids is comma-separated, max500; returns only requested members. An absent requested id is deleted—drop it locally. Coordinates: board units, top-left origin, +x right/+y down. Example: get_canvas_elements({id:"<boardId>", ids:"el-a, el-b"}) returns only surviving elements. HTML embed markup appears as html only for one requested id; multi-id, whole-board and manifest reads omit markup. With canvas_embeds off, the single HTML read returns 404 module_unavailable; ordinary element reads omit html.
Use to create, patch or delete several elements in one call (building a board, laying out a diagram, moving or deleting a group) instead of repeated create_canvas_element, patch_canvas_element or delete_canvas_element calls. operations run in order: create {element: the create_canvas_element body}, patch {element_id, patch: the patch_canvas_element body}, delete {element_id}. Each succeeds or fails alone with the reason its single-element tool gives; failures leave the rest applied. A create's temp_id works later wherever an element id goes (element_id, frameId, connector from/to element); temp_ids maps it to the minted id. Too many operations: 400 canvas_batch_too_large names the limit; split. Coordinates: board units, top-left origin, +x right/+y down. Example: batch_canvas_elements({id:"<boardId>", operations:[{op:"create", temp_id:"a", element:{type:"sticky", x:0, y:0, w:160, h:120, content:"Idea"}}, {op:"create", temp_id:"b", element:{type:"shape", shape:"rect", x:240, y:0, w:160, h:120}}, {op:"create", element:{type:"connector", from:{element:"a", anchor:"e"}, to:{element:"b", anchor:"w"}, x:0, y:0, w:0, h:0}}]}).
Use to create a new sticky/text/shape/image/codeBlock/frame/swimlane/connector/metric/embed/portal. Several elements: one batch_canvas_elements call writes the board once; repeated single calls each wait on the board lock. Fields are flat, id is the board; server mints element id and preserves others. Coordinates: board units: top-left origin, +x right/+y down; x, y, w, h geometry. Type fields: sticky/text content; shape shape/content; image src; codeBlock code/language; frame name/metricWindow; metric report insights:<id>|funnels:<id>, optional step/title/window {from, to} (today-28d/now), live report values. Connector from/to: {element, anchor:n|s|e|w} follows shapes; {point:{x, y}} stays fixed; x, y, w, h=0. Example: create_canvas_element({id:"<boardId>", type:"sticky", x:100, y:80, w:160, h:120, content:"Idea"}). Embed: sandboxed frame, kind html/url, title, interactive=false (no scripts/forms); true allows interaction. html for kind html or src for kind url; https URL required (400 embed_url_invalid). Never send src for html. HTML: inlined single file, 5 MB UTF-8 max (413 embed_html_too_large), stored board asset/src. Responses/board reads: src, never markup; only get_canvas_elements with one id returns html. No markup messages; comments reach you by mention or reading the board, never wake you unmentioned. Embed example: create_canvas_element({id:"<boardId>", type:"embed", kind:"html", title:"Mockup", interactive:false, x:0, y:0, w:640, h:480, html:"<!doctype html><button>Pay</button>"}). canvas_embeds off: create, html/src/interactive patches, URL mint, serve, both asset markup reads 404 module_unavailable; single read omits html; move/resize/rename/lock/delete work. Portal: live targetId as viewer; modules/selectors/default sizes docs {}/480x640, slides {slide} zero-based first default/480x270, sheets {range} A1 import-range cap/560x320, canvas {frameId}/560x400. Wrong-module/malformed/over-cap 400 portal_selector_invalid; inaccessible target module 404; title server-written create/patch, never send. Host board/portal-containing frame 400 portal_recursive; connection 400 portal_not_connectable, use create. Geometry required. Example: create_canvas_element({id:"<boardId>", type:"portal", module:"sheets", targetId:"<id>", selector:{range:"Sheet1!A1:D10"}, x:0, y:0, w:560, h:320}). canvas_portals off: create/selector patch 404 module_unavailable.
Use to extend a flowchart/map/swimlane: source_element_id and direction n/s/e/w place a new node one standard gap away, clear overlaps, and create its attached connector. Optional drop_point {x, y} centers the node there instead. element sets its type; omit to duplicate source. Connector/frame/swimlane types refused; portal source returns 400 portal_not_connectable. Coordinates: board units, top-left origin, +x right/+y down. Example: add_connected_element({id:"<boardId>", source_element_id:"<elId>", direction:"e", element:{type:"shape", shape:"diamond"}}) returns element and connector.
Use to patch one element to move/restyle/relabel/reanchor: only changed fields; id/type fixed, others preserved. Several elements: one batch_canvas_elements call writes the board once; repeated single calls each wait on the board lock. Style replaces; style:null removes. Coordinates: board units: top-left origin, +x right/+y down; x, y move top-left. Example: patch_canvas_element({id:"<boardId>", elId:"<elId>", x:240, y:80}). Embed: sandboxed frame; interactive=false runs no scripts/forms; kind fixed; title/interactive editable. HTML single inlined UTF-8 file, 5 MB max (413 embed_html_too_large); new board asset/src, old markup in version history. src only URL kind, https required (400 embed_url_invalid otherwise; HTML src returns 400 embed_src_managed). Responses src, never markup; only get_canvas_elements with one id returns html. Embed example: patch_canvas_element({id:"<boardId>", elId:"<elId>", html:"<!doctype html><button>Pay</button>"}). canvas_embeds off: create, html/src/interactive patch, URL mint, serve, both asset markup reads 404 module_unavailable; single read omits html; move/resize/rename/lock/delete work. Portal module/targetId fixed; patch selector: docs {}, slides {slide} zero-based first default, sheets {range} A1 import-range cap, canvas {frameId}. Wrong-module/malformed/over-cap 400 portal_selector_invalid. Viewer access, target module 404; title server-written create/patch, never send. Host/portal-containing frame 400 portal_recursive; connection 400 portal_not_connectable. Resize docs480x640, slides480x270, sheets560x320, canvas560x400. Example: patch_canvas_element({id:"<boardId>", elId:"<elId>", selector:{range:"Sheet1!A1:F20"}, w:560, h:320}). canvas_portals off:selector patch 404 module_unavailable; move/resize/lock/delete work.
Delete one element from a board by id. Use when removing an object; connectors attached to it are not auto-deleted, so patch or delete those separately. A locked element answers 409 element_locked; unlock it first. Several elements: one batch_canvas_elements call writes the board once; repeated single calls each wait on the board lock. Every other element is untouched. Coordinates: board units, top-left origin, +x right / +y down (unchanged by a delete). Example: delete_canvas_element({ id: "<boardId>", elId: "<elId>" }).
Read the workspace People lifecycle registry, including archived stages. An unconfigured workspace uses the seven default stages.
Add a People lifecycle stage with an immutable key. Workspace admins only. Keys cannot be deleted or reused; archive a stage to hide it from assignment controls.
Rename, reorder, archive or restore a People lifecycle stage without changing its stable key or contacts. Workspace admins only.
List all users (humans and agents) in the current workspace who can own CRM records. Use before assigning contacts or deals to a specific owner_id.
List the caller's address book: the shared CRM contacts they can see unioned with their own privately captured addresses, deduplicated on email with the shared record winning. Use this to resolve a recipient the user has written to before instead of asking them to retype an address. Private entries are never visible to another member.
List the workspace's typed custom-property definitions for contacts and deals, including relation properties that link a record to another contact, company, or deal. Optionally filter by entity_type.
Define a typed custom property for contacts or deals. field_type 'relation' (with relation_target contact|company|deal) makes the property a link to another record: values written to custom_properties under this key must be the id of an existing record of the target type.
Update a custom-property definition's label or select options.
Delete a custom-property definition. Existing custom_properties values are left in place; they simply stop being typed.
Get pipeline metrics summary: total open deals, total deal value, win rate, deals won/lost, and recent activity counts. Useful for high-level pipeline health questions.
Get CRM revenue, deals-by-source, pipeline-by-owner, and activity reports for accessible records, optionally scoped to one pipeline. Each report preserves its own timestamp axis.
List CRM contacts. Supports filtering by type, lifecycle stage, owner, and text search. Returns contacts with id, name, email, phone, type, lifecycle_stage, company, and timestamps. Use this when the user asks to see their contacts, find a person, or list companies. Also use this with the 'q' parameter to search for existing contacts BEFORE creating new ones to prevent duplicates.
Create a new CRM contact (person or company). Persons can be linked to a company via company_id. IMPORTANT: Before creating a contact, ALWAYS call list_contacts with a search query (q) to check if a contact with the same name or email already exists. If a matching contact is found, use update_contact instead of creating a duplicate. Only create a new contact if no match is found. Use this when the user asks to add a contact, log a lead, or create a company record.
Fetch multiple CRM contacts by their IDs in a single request (max 100). Missing or inaccessible IDs are silently omitted from the result.
Merge two CRM contacts that represent the same person/company. The secondary contact is deleted, and its activities and deals are re-linked to the primary. Use this when you discover duplicates (same email or name). The primary contact is the one kept; the secondary is removed.
Delete many CRM contacts in a single operation. Returns counts of deleted/skipped/failed items.
Update many CRM contacts in a single operation. Useful for re-assigning ownership, updating lifecycle stage on a list of contacts, or reclassifying industry. Provide an array of contact IDs and the fields to apply to all of them.
Export selected contacts as CSV or vCard. CSV includes core fields and custom properties; vCard preserves imported RFC 6350 properties and parameters.
Get a single CRM contact by ID, including full details: name, email, phone, title, company, lifecycle stage, custom properties, and timestamps. If the user is currently viewing a contact, use the contact ID from the page context.
Update an existing CRM contact's fields. Provide only the fields you want to change. For interrupted sync recovery, read the contact and send expected_updated_at with only remaining changes. A 409 carrying contact_id means another contact already has that email; any other 409 means the contact changed: read again and reconcile; never replay an earlier snapshot over newer values. If the user is currently viewing a contact, use the contact ID from the page context.
Delete a CRM contact (owner-scoped). Permanently removes the contact record. Use with caution — prefer updating lifecycle_stage to 'evangelist' or archiving via merge instead.
List activities (notes, calls, emails, meetings, tasks) on a specific contact's timeline. Useful for summarizing recent interactions with a contact before sending an email or making a call.
Get a contact's current lead score with a breakdown of which scoring rules matched and how many points each contributed. Use this to explain why a contact is hot/cold.
List files attached to a contact
Copy a readable Drive file into a contact-owned attachment. The copy is readable by the contact readers; the original file and its sharing remain unchanged. A native doc, sheet, deck, canvas or form without stored bytes returns 422 `native_document_not_copyable`; attach it as a link.
Detach a file from a contact
Serve the photo image stored for a CRM contact.
Delete a CRM contact's profile photo. Clears the avatar_url and removes the stored image.
List all contact groups in the workspace with their member counts. Groups are named collections for organizing contacts (e.g., 'Conference leads 2026').
Create a named contact group for organizing contacts. Groups allow overlapping membership — a contact can be in multiple groups.
Update an existing contact group's name, color, or sort order. Group name must remain unique within the workspace.
Delete a contact group. Memberships are removed but contacts themselves are preserved.
List all contacts that are members of a contact group. Returns the contact id, name, email, and type for each member.
Bulk add or remove contacts from a contact group. Idempotent — adding an already-present contact or removing an absent one is a no-op.
List CRM deals. Supports filtering by pipeline, stage, and status. Returns deals with id, title, amount, currency, stage, pipeline, contact, status, and timestamps. Use this when the user asks to see their deals, check the pipeline, or find opportunities.
Create a new CRM deal (revenue opportunity). Deals belong to a pipeline and sit in a specific stage. Link to a contact and/or company. Use this when the user asks to create a deal, log an opportunity, or start tracking a sale.
Fetch multiple CRM deals by their IDs in a single request (max 100). Missing or inaccessible IDs are silently omitted from the result.
Get a single CRM deal by ID, including full details: title, amount, currency, stage, pipeline, contact, company, owner, probability, close date, custom properties, and timestamps. Pass embed=line_items to include the full line-items array. If the user is currently viewing a deal, use the deal ID from the page context.
Update an existing CRM deal's fields. Provide only the fields you want to change. Use this to move a deal to a new stage, update the amount, mark as won/lost, etc. If the user is currently viewing a deal, use the deal ID from the page context.
Delete a CRM deal (owner-scoped). Permanently removes the deal record and its line items. Prefer updating status to 'lost' with a lost_reason for reporting purposes.
List activities for deal (owner-scoped)
Delete many CRM deals at once by their ids.
Update many CRM deals in a single operation. Useful for moving multiple deals to a new stage, reassigning ownership, or marking a batch of deals as won/lost. Provide an array of deal IDs and the fields to apply.
Export selected deals as CSV
List all line items for a deal, plus the computed total.
Add a line item to a deal. Either reference an existing product (via product_id) or create a custom line by providing name and unit_price. The deal amount is auto-recalculated.
Update a deal line item (quantity, price, discount, name/description). Amount auto-recomputed. If sync_deal_amount defaults true, the deal's total is also refreshed.
Delete a deal line item. By default the deal's total amount is re-synced after deletion.
List files attached to a deal
Copy a readable Drive file into a deal-owned attachment. The copy is readable by the deal readers; the original file and its sharing remain unchanged. A native doc, sheet, deck, canvas or form without stored bytes returns 422 `native_document_not_copyable`; attach it as a link.
Detach a file from a deal
List deal templates in the workspace (reusable deal presets: default pipeline, stage, amount, probability, and line items).
Get a single deal template with its line items and resolved pipeline/stage names.
Create a deal template — a reusable preset for creating deals with standard pipeline/stage/amount and line items. Use when the user has a repeatable sale like 'Standard Enterprise Plan'.
Update a deal template. Providing `line_items` replaces the entire line-item list.
Delete a deal template (does not affect deals previously created from it).
Create a new deal from a saved deal template. The template provides default pipeline, stage, amount, probability, and line items. Use this when starting a deal that matches a known pattern (e.g., 'Standard Enterprise Plan').
List all CRM pipelines with their stages. Each pipeline contains an ordered list of stages with names, sort order, and default probabilities. Use this before creating or moving deals to find valid pipeline_id and stage_id values.
Create a new CRM pipeline with ordered stages. Each stage has a name and an optional default probability (0-100). Most workspaces need only one pipeline; create additional ones for distinct sales processes (e.g., 'New Business' vs 'Renewals').
Get a single CRM pipeline by ID with its ordered stages and its `visibility` — `workspace` when the pipeline is shared with the whole workspace, `private` when only its members reach its deals. A caller who is not a member receives 404.
Update a pipeline's name or stages. Providing `stages` replaces the entire stage list — include every stage you want to keep. Reordering is done by the order in the array.
Delete a CRM pipeline and all of its stages. Deals in the pipeline must be moved or deleted first.
Get analytics for accessible deals, optionally scoped to one pipeline: pipeline value by stage, conversion rates, average deal size, sales velocity, and deals breakdown.
List the members of a CRM pipeline. Deal visibility is pipeline membership: a deal is visible to admins/owners and to members of its pipeline (directly or via a team). Roles are admin (can manage the pipeline + members) or member.
Add (or update the role of) a user or team on a CRM pipeline. Provide either user_id or team_id, plus role (admin | member). Admins/owners and pipeline admins can manage members.
Remove a user or team from a CRM pipeline. The principalId is the user_id or team_id of the member to remove.
List recent CRM activity visible to the current user across contact and deal timelines.
Log an activity on a CRM contact or deal timeline. Activities track interactions like calls, emails, meetings, notes, and tasks. Use this when the user asks to log a call, record meeting notes, or add a note to a contact/deal.
Get a single activity by ID. Useful for reading back a just-created activity via the resource URL in the activity.created event payload.
Delete one logged CRM activity by id.
List all CRM email/task sequences (multi-step automated outreach campaigns). Returns sequences with their steps, active status, and current enrollment count.
Create a multi-step CRM sequence (email/wait/task/call/branch steps). Use this when the user asks to build a nurture campaign or outreach cadence.
Get a single sequence with its full step list and current enrollment count.
Delete a sequence (and all its enrollments).
List all enrollments for a sequence, including each contact's current step, status, and next execute time. Useful for checking progress or finding an enrollment ID to unenroll.
Permanently remove a contact from a CRM sequence (terminal — sets status to 'unenrolled'). The contact will not receive any further sequence steps. Returns 409 if already in a terminal status. Provide the enrollment ID (not the contact ID — find it via list_sequence_enrollments).
Get the event log for a single sequence enrollment (step-executed, email-opened, email-replied, branch-taken, completed). Useful for auditing why a contact moved to a particular step.
List the contact fields that can be used in scoring rules (lifecycle_stage, industry, title, website, custom_properties.*, etc.) and the operators each supports.
Recalculate lead scores for all contacts in the workspace. Use this after creating or updating scoring rules so existing contacts pick up the new scoring.
List all lead scoring rules for the workspace.
Create a lead scoring rule. When a contact matches (field + operator + value), the specified points are added to their lead score. Positive points indicate engagement; negative points indicate disqualification signals.
Update a lead scoring rule (name, field, operator, value, points, is_active).
Delete a lead scoring rule.
List CRM quotes/proposals. Filter by status (draft, sent, accepted, rejected, expired), deal, or contact. Useful for finding outstanding quotes or recently-accepted ones.
Create a CRM quote (proposal) with line items. Auto-computes subtotal, tax, discount, and total. Link to a deal/contact/company via IDs. Status defaults to 'draft' — use send_quote to email it.
Get a single CRM quote with its line items and linked deal/contact/company/owner.
Update a CRM quote. Providing `line_items` replaces the entire line-item list. Totals are auto-recomputed.
Delete a CRM quote and its line items.
Send a CRM quote via email to the contact. Marks the quote as 'sent'. By default, sends to the linked contact's email address.
Export quote as PDF (or printable HTML with ?format=html)
List products in the workspace catalog. Use this when the user asks about pricing or available SKUs, or before adding a product as a deal line item.
Create a product in the workspace catalog. Products are the reusable items that can be attached to deal line items and quote line items. Set is_active=false to retire without deleting.
Get one CRM product from the catalog by id.
Update a CRM product's name, price or other fields.
Delete a product from the workspace catalog. Existing deal/quote line items referencing it are NOT deleted, but their product_id is nulled.
List the user's saved CRM views (filter + sort presets for contacts/companies/deals). Views personalize how CRM data is displayed without changing the underlying records.
Create a saved CRM view: a named filter+sort preset for contacts, companies, or deals. Useful when the user asks to 'save this filter' or 'create a view for high-value deals'.
Update a saved CRM view (rename, change filters, toggle default/shared).
Delete a saved CRM view.
List the user's scheduler links (Calendly-style booking pages). Each link has a shareable slug, duration, availability hours, and optionally auto-creates a contact + deal when booked.
Create a scheduler link (booking page). Returns the public URL slug and availability settings. If auto_create_contact is true, every booking auto-creates a CRM contact + deal.
Get a single scheduler link (owner-scoped) with its settings and booking count.
Update a scheduler link (rename, change availability, toggle active, etc.).
Delete a scheduler link. Existing confirmed bookings remain but the public booking page stops accepting new bookings.
List all bookings for a single scheduler link (owner-scoped).
List all bookings (across all scheduler links) owned by the current user. Useful for summarizing the user's upcoming meetings.
Cancel a booking as its host. The booked event is cancelled and the booker is always emailed the cancellation, with the optional message.
List the user's forms. Returns forms with id, title, slug, is_published, response count, timestamps, and the caller's own star and Drive tags. Pass `starred=true` for only the forms you starred. Use this when the user asks to see their forms, find a specific form, or check form status.
Create a new form for collecting data (surveys, feedback, registrations, quizzes). Returns the created form with id and public slug. Use this when the user asks to create a survey, questionnaire, feedback form, registration form, or any structured data collection. Define fields with types, labels, validation, and optional conditional logic. is_published=true immediately opens the responder link to public responses.
List responses submitted to a form. Returns responses with respondent info, submitted data (keyed by field ID), and timestamps. Use this when the user asks to see form results, check submissions, analyze feedback, or export response data.
Download all matching form responses as CSV with field labels, display values, blank unanswered cells, and ISO submission timestamps. Streams beyond the response-list page size.
Get a single form response by ID. Returns respondent info, submitted data (keyed by field ID), and timestamp. Use when the user asks about a specific submission or wants to review a particular response.
Delete a single form response. This is destructive — always confirm with the user before executing. Use when the user asks to remove or delete a specific submission.
List A/B test variants for a form. Returns each variant's id, name, traffic weight, active status, and variant form ID. Use when the user asks about their A/B tests, variant list, or experiment setup.
Create a new A/B test variant by duplicating the parent form. Optionally override fields to change question wording, field types, options, or validation. The variant inherits the parent's publish state. Traffic weight (0–100) controls the percentage of visitors who see this variant versus the control. Use when the user asks to set up an A/B test or add a variant.
Update an A/B test variant's name, traffic weight, or active status. Use when the user asks to rename a variant, adjust the traffic split, or pause/resume an experiment.
Delete an A/B test variant (and its underlying duplicated form). Destructive — always confirm with the user. Use when the user asks to remove a variant or end an experiment.
Get analytics for a form: total views, total responses, completion rate, average completion time, responses-by-day (last 30 days), per-field drop-off funnel, and per-field summaries (option counts for select fields, numeric stats for rating/number fields). Use when the user asks about form performance, response rate, completion rate, or field-level statistics.
Get A/B test comparison analytics for a form with variants. Returns control + each variant with views, responses, completion rate, and traffic weight. Use when the user asks which variant is winning, for A/B test results, or experiment performance comparison.
Publish a form so it accepts public responses. Semantic shortcut for PATCH /api/forms/{id} with is_published=true; same validation (every field needs a non-empty label).
Unpublish a form so it stops accepting public responses. Semantic shortcut for PATCH /api/forms/{id} with is_published=false.
Get a form's full details including field schema, settings, publish status, and slug. Use this when the user asks about a form's fields, structure, or configuration. If the user is currently viewing a form, use the resource ID from page context. is_published changes public response availability; fields replaces the question list.
Update an existing form's title, description, fields, settings, or publish status. Use this when the user asks to edit a form, add/remove fields, publish/unpublish, or change form settings. If the user is currently viewing a form, use the resource ID from page context. is_published changes public response availability; fields replaces the question list.
Delete a form and all its responses. This is a destructive action — always confirm with the user before executing. Use this when the user explicitly asks to delete or remove a form.
Get the available color palette for calendars and events. Returns 24 calendar colors and 11 event colors, each with a background and foreground hex value. Use this when the user asks what colors are available, or before setting a calendar or event color.
Check free/busy availability for one or more users in a given time range. Returns busy time slots (not event details) for each user. Use this when the user asks to find a free time, check if someone is available, or schedule a meeting. Call this before create_event to avoid scheduling conflicts.
Detect attendee scheduling conflicts.
Check out-of-office status for one or more users. Returns per-user OOO flag and event details when applicable. Call this before scheduling to avoid inviting people who are OOO.
Resolve an attendee identifier to a current workspace member without changing the event or exposing foreign identities.
Reminders due to fire within the next `window_hours` (default 24, max 168), soonest first. Each row carries the event it is about — title, span, status, location and calendar — so the caller renders the row without a second fetch.
List the user's calendars, creating their default calendar if none exists. Returns calendar id, name, color, timezone, and whether it's the default. Use this to find the user's default calendar ID before creating events.
Create a calendar in the current workspace.
Start external provider OAuth.
List connected external accounts.
Disconnect external account.
List calendars on connection.
Enable two-way sync.
Schedule an immediate background pull for this connection. A 202 response accepts the request; it does not report completion. Poll the connection's sync_state for the returned request_id until completed or failed. last_sync_at can advance during partial failure and is not a completion signal. Concurrent requests share the active request_id.
Disable sync for one calendar.
List calendar events across all of the user's calendars. Supports filtering by date range, calendar, and free-text search. Returns events with id, title, start/end times, location, status, and attendees. Use this when the user asks to see their schedule, upcoming meetings, or events for a date. Use the q parameter to search for events by keyword (matches title, description, location, and attendee names).
List events scoped to a single calendar by its id — the RESTful path variant of list_events (`/api/calendars/events?calendar_id=…`). Requires at least viewer access to the calendar.
Full-text search across all calendar events for the current workspace. Returns matching events with id, title, start/end times, and calendar info. Use this when the user asks to find a specific event by name or description keyword.
Create a new calendar event. Requires a title, start time, and end time. Optionally add a description, location, attendees, recurrence, and more. Use this when the user asks to schedule a meeting, add an event, or book time. If the user doesn't specify a calendar, use list_calendars to find their default calendar ID first.
Get a single calendar event by ID, including full details: title, description, location, start/end times, recurrence rule, status, visibility, and attendees with RSVP status. For a recurring master, occurrence_start_at selects an occurrence by its original recurrence instant, retaining a saved occurrence's own ID after rescheduling. If the user is currently viewing an event, use the event ID from the page context.
List individual occurrences of a recurring event within a date range. Expands the RRULE server-side and returns instance objects with computed start/end times. Use this to find specific occurrences (e.g., 'what meetings are on next Tuesday?') or to list upcoming instances of a series.
Update an existing calendar event. Provide only the fields you want to change. Use this to reschedule, change attendees, update the description, or cancel an event. If the user is currently viewing an event, use the event ID from the page context.
Delete a calendar event. This is a destructive action — always confirm with the user before calling. If the user is currently viewing an event, use the event ID from the page context. Guests are emailed a cancellation only when `send_updates` asks for it.
RSVP to a calendar event that the user is invited to. Use this when the user asks to accept, decline, or mark tentative on an event invitation.
Export single event as .ics.
Edit a single occurrence of a recurring event, creating an exception without affecting other occurrences. Use this when the user says 'move just next Tuesday's meeting to 3pm' or 'cancel only this one occurrence'.
Edit this and all future occurrences of a recurring event by splitting the series at the given date. Use this for 'change the meeting time from next week onwards' or 'update our recurring Friday sync starting next month'.
Edit all occurrences in a recurring series. Updates the master event.
Delete a single occurrence of a recurring event by adding the occurrence date to the master's excluded_dates. The occurrence is suppressed without affecting other instances.
Delete this and all future occurrences of a recurring event by adding an UNTIL clause to the master's RRULE, ending the series the day before the given split_date. When no occurrence starts before split_date, the whole series (master and every exception) is deleted instead.
Book one or more resources (rooms, equipment) for an existing event. Returns a conflict error if a resource is already booked.
Release every resource booked for a calendar event.
List resources currently booked for an event (rooms, equipment).
List event reminders.
Add a reminder to a calendar event.
Replace reminder list.
Create a meeting-notes doc for an event, or — when `document_id` is given — link that existing doc and grant attendees access.
Unlink meeting-notes doc.
List bookable resources (rooms, equipment) in the workspace. Call before booking a room to see what's available.
Create a bookable resource.
Get one bookable calendar resource by id.
Update a bookable calendar resource's details.
Delete a bookable calendar resource by id.
Check if a specific resource (room, equipment) is free in a time range. Call before booking to avoid conflicts.
Delete one calendar event reminder by id.
Get a single user's permission entry on a calendar, including their role, display name, and email. Useful for checking if a specific user has access without listing all permissions.
Rotate a published calendar's read-only subscription URL. The previous URL is revoked immediately.
Revoke the public .ics feed for a calendar, making it private again.
Import an .ics feed by URL. Fetches the URL server-side and imports its events into the calendar.
Export entire calendar as .ics.
Delete ALL events from a calendar. Owner-only — use for resetting shared calendars at quarter boundaries or wiping test data. Returns the count of deleted events. This is irreversible.
Get a single calendar by ID. Returns the calendar's name, color, timezone, default status, and owner. Use this when the user asks about a specific calendar's settings or before creating events to verify the timezone.
Update calendar fields.
Delete a calendar and every event on it.
Export a separately labeled, digest-bound native Wiki backup containing the workspace Wiki graph and page-owned image metadata and bytes. This is not DOCX.
Restore a native Wiki backup into its empty source workspace. Preserves spaces, hierarchy, page content, databases, backlinks, synced blocks, permissions, versions, comments, and v2 page-owned images. Historical v1 backups contain only the graph.
List all wiki spaces. Returns spaces with id, name, slug, description, icon, owner info, privacy setting, and page count. Use this when the user asks to see wiki spaces, browse the knowledge base, or when you need a space ID to create a page.
Create a new wiki space (a top-level container for pages). Use this when the user asks to create a new knowledge base area, team wiki, or documentation section. Spaces organize pages by team or domain (e.g., 'Engineering', 'Product', 'Company Handbook').
List distinct members of a space — the owner plus everyone with a page-level permission inside it.
Add a user as a direct member of a wiki space with a given role.
Remove a direct member from a wiki space.
Get backlink graph data for a wiki space. Returns nodes (pages) and edges (links between pages). Use this when the user asks to visualize the knowledge graph, see how pages connect, or find highly-linked 'hub' pages.
List pages in a wiki space with offset/limit pagination. Returns page summaries (id, title, slug, icon, parent_page_id, sort_order). Use this when the user asks to see all pages in a space, or when you need to paginate through a large space.
Create a new wiki page within a space. Pages use Markdown content and support nested hierarchies via parent_page_id. Use this when the user asks to create a wiki page, add documentation, write a runbook, or add to the knowledge base. Call list_wiki_spaces first to get the space ID.
Get a wiki space by slug, with its full page tree. Supports cursor pagination on `pages`.
Update a wiki space. Only the space owner may change settings.
Delete a wiki space and all of its pages.
List trashed (soft-deleted) wiki pages in the workspace.
Restore a trashed wiki page and its children.
List the immediate child pages of a wiki page. Use this to traverse the page tree one level at a time, or to count children before a destructive delete.
List wiki pages that contain `[[link]]` references to this page. Use this to see what pages cite a given page, or to find inbound knowledge links.
Add an image to wiki body/cover. Returns the parent-owned image URL `/api/assets/{id}/content`. Input: content_base64 bytes, source_file_id (copies a viewable Drive file), or source_url (copies public HTTPS through SSRF guard); no source link retained. Native sources without stored bytes return 422 native_document_not_copyable; attach it as a link. Byte-sniffed images only, 415 for other types or malformed SVG. SVG sanitized on upload and every read, inline image/svg+xml with restrictive CSP; malformed stored SVG returns 404. Drive SVG stays attachment. Owner quota exceeded 413; same image deduplicated per parent. Absent from Drive, readable exactly with parent.
List the saved version snapshots for a wiki page (most recent first). Use this when the user asks for revision history, wants to see who edited and when, or needs to find a version to revert to.
Create a manual snapshot of the current wiki page state. Useful before a risky edit or when the user wants to mark a 'known good' checkpoint. Optional `label` names the snapshot. Returns the new version's id + metadata.
Get a wiki page version's full content (title + markdown body). Use this when the user asks what a previous version looked like, wants to compare versions, or needs the content of a specific snapshot.
Revert a wiki page to a previous version snapshot. This replaces the current content with the version's content and rebuilds the page's collaborative state from it. A snapshot of the current state is automatically created before reverting. This is destructive for the in-progress edit — always confirm with the user first.
Duplicate a wiki page (and optionally its descendants) into a target space + parent. Returns the new top-level page.
Export a wiki page to docx, pdf, html, or md. Streams the file with a Content-Disposition attachment header.
List all inline databases hosted on a wiki page.
Create an inline database on a specific wiki page.
Move the caller's project-less tasks linked from this wiki page into `project_id`; set the page's tasks project if unset. Other linked tasks are listed in `skipped` and unchanged.
Get a wiki page's full details including Markdown content, child pages, backlinks (pages that link to this page), and metadata. Use this when the user asks about a page's content, wants to read documentation, or needs to check what links to a specific page. If the user is currently viewing a wiki page, use the resource ID from page context.
Update an existing wiki page's content, title, icon, or parent. Use this when the user asks to edit a wiki page, update documentation, or move a page to a different parent. If the user is currently viewing a wiki page, use the resource ID from page context.
Move a wiki page to Trash, together with every page beneath it. Trashed pages leave the tree and search, and can be brought back with `wiki_page_restore`; only `wiki_page_permanent_delete` removes them for good. A page that has child pages answers 409 with `requires_cascade` and `children_count` unless `cascade=true` is passed, so confirm with the user before trashing a sub-tree. Use this when the user asks to delete or remove a wiki page.
Case-insensitive substring search across wiki page titles and Markdown content. Returns only pages the caller can view under the effective page-to-space permission rules, with id, title, space, and content snippets. Use this when the user asks to find information in the wiki, look up documentation, or search the knowledge base. More efficient than reading individual pages.
Create an inline database on a wiki page. Requires editor role on the host page.
Append a row to an inline database. Requires editor role on the host page. Set a person property to a user id from the users list; other text is stored and shown as written.
Update properties or sort order on a database row. Properties merge: named properties change, others stay, and null clears one. Set a person property to a user id from the users list; other text is stored and shown as written.
Delete a database row.
Get a single database row by id.
Query rows from an inline database with optional view, filter, and sort. Returns paginated results.
Auto-create the reverse relation property on the target database if missing. Idempotent — returns the existing or newly-created reverse property id.
Compute and return rollup values for all rows in the database. Each rollup aggregates a property across related rows.
Get an inline database with its full row set.
Update an inline database's name, schema, or views. Requires editor role on the host page.
List synced blocks that were originally created on a specific page. Use this when the user asks what shared content a page publishes, or to audit reuse.
Create a synced block — a reusable content fragment that can be embedded in multiple wiki pages and stays in sync everywhere it's used. Returns the new block's id. Use this when the user asks to create a shared snippet, reusable callout, or cross-page content.
Get a synced block by ID. Returns the block's content and metadata. Use this when you need to read the current content of a shared block.
Update a synced block's content. All pages embedding this block will reflect the new content. Use this when the user asks to edit shared content or update a reusable snippet.
Soft-delete a synced block. Embedded references will show error chips until restored. Use restore_wiki_synced_block to undo.
Restore a previously deleted synced block. Embedding pages will render the block again instead of showing error chips.
Get parent-owned image bytes for doc/sheet/deck/board/wiki/task, absent from Drive. Readable exactly by parent role/access, including external collaborators. Proxied, no redirect; Cache-Control:no-store. Active types including SVG and Canvas HTML download as attachments. With canvas_embeds off, HTML returns 404 module_unavailable; missing, inaccessible or invalid assets return 404. Transient storage failures return 503 asset_storage_unavailable with Retry-After.
Get the current workspace details (id, name, slug). Requires admin role.
Update workspace name or slug. Requires admin role.
Remove workspace logo (admin only)
Get all workspace settings. Requires admin role.
Update workspace settings (name, logo, timezone, etc.). Requires admin role. Pass a flat key-value object.
Get workspace security settings (require_2fa, session_timeout_hours, password_min_length, ip_allowlist). Requires admin role.
Read workspace-default email notification prefs. Returns sensible defaults (all enabled) when never configured.
Partially update workspace-default email notification prefs. Emits settings.updated audit event.
Get the workspace default email signature (Markdown). Requires admin role.
Set the workspace default email signature. Pass null (or omit markdown) to clear. Requires admin role.
Get email deliverability statistics (sent/bounced/complained/opened rates) for the past 30 days. Requires admin role.
List workspace members and agents with identity, role, status, storage usage, and timestamps. Requires admin or owner.
Get one workspace user's full admin record by id.
Update a member's display name, username, avatar, storage quota, or workload capacity. Requires admin or owner. Change roles with assign_role/remove_role; suspend or restore access with admin_user_suspend/admin_user_unsuspend.
Read a deleted member's durable erasure outcome and notice-delivery state. Requires admin or owner.
Email the member a one-hour password-reset link. When the send fails, `sent` is false with `error: "delivery_failed"`. The link itself is returned only to a REST caller, never to an AI tool surface.
List pending, accepted, and expired workspace invites, pending first and then newest. Paginated; `pending_total` counts invites awaiting acceptance. Each row also carries `delivery` — what the provider said about the latest send of that invite email, so a bounced invite is distinguishable from an ignored one.
List all roles in the workspace (system roles + custom roles). Requires admin role.
Get workspace usage statistics including storage breakdown by user and module, active user count, API call volume, and per-module adoption metrics. Requires admin or owner role. Use this when the user asks about workspace usage, storage consumption, billing metrics, or capacity planning.
Daily creation counts (last 14 days, split into "last 7" + "prior 7" windows) for the four hero metrics on the Overview tab. Returns `{users, documents, messages, emails}` each with `{buckets[7], total, previous_total}` so the UI can draw a sparkline + render a 7...
Export usage stats as CSV
Trigger DNS verification for a workspace domain. Requires admin role.
Verify custom domain ownership via DNS TXT record. On success, auto-registers with SES to retrieve DKIM/SPF records. Requires admin role.
Send a test notification email (useful for checking email rendering and deliverability). Requires admin role.
list suppressed addresses
manually add a suppression
Clear a bounce once the address is fixed. An unsubscribe or spam complaint is lifted only by the recipient, so clearing one is refused with 409 suppression_recipient_controlled; clearing a bounce on an address that also unsubscribed keeps the unsubscribe.
list all integrations for workspace
list Slack channels via bot token
update integration config
disconnect an integration
The caller's IMAP, SMTP-submission and CalDAV settings as data — the same username, hosts and ports the `.mobileconfig` encodes, for a mail client no Apple profile covers. Hosts come from `@ambiguous/topology`, so no client carries a hostname literal. Gated exactly like the profile: 400 when the account has no Ambiguous password or no resolvable workspace address, because either makes the settings unusable.
Get the active workspace (name, slug, logo, scheduling timezone). Accessible to any authenticated user.
Check whether a workspace slug is available.
List workspaces the current user belongs to (human users may have multiple). Returns workspace id, name, slug, role, and which one is active.
List event types an agent can subscribe to. Includes the `*` wildcard.
List webhook registrations owned by the caller.
Get one webhook subscription's config by id.
Delete one webhook subscription by id.
List recent delivery attempts (status, response, retries). Paginated via `?limit=` (default 25) and `?offset=` (default 0).
List notifications for the current user. Supports cursor pagination via `?cursor=`; legacy offset pagination still works. Filters: `resource_types`, `excluded_resource_types`, `event_types`, `unread_only`; visibility toggles: `include_dismissed`, `include_snoozed`.
List active mute rules. Pass `?include_expired=true` to also see rules whose `until` has passed (these no longer suppress notifications).
Create or refresh your in-app/live notification suppression rule for resourceId, eventType, or both. Matching direct mentions are suppressed there too; email and push delivery are unchanged. This does not change a chat channel's mute setting or notification preference. For ordinary channel mute/unmute use set_channel_muted. At least one selector is required; resourceId also requires resourceType. until=null or omitted means indefinite. A matching rule's expiry is refreshed.
Remove a mute rule by id. Scoped to the owning user.
Mark every unread notification in the optional resource/event scope as read.
Mark a single notification read. Returns `was_unread` — true only if THIS call flipped it. A consumer that gets `false` skips the work, so two processes on one identity cannot both act on one notification.
Convenience alias for `PATCH /api/notifications/:id` with `{read:false}`.
Soft-delete (hide from default feed). Preserved for audit / undismiss.
Restore a dismissed notification.
Hide until `{until: ISO8601}`. Pass `{until: null}` to clear snooze.
OCR a photo of a business card and extract structured contact fields (name, title, company, email, phone, website, address, notes). Returns JSON with each field nullable when not present on the card. Use when the user uploads or references a business card image and wants to add the contact to CRM. The image must be passed as a base64 data URL (data:image/jpeg;base64,...) of an image format Gemini Vision supports (JPEG, PNG, WebP, HEIC). Max 6 MB.
Get the full unread projection (mail inbox, per-channel chat badges with mute/preference resolved, notifications, total). Used for cold-connect / reconnect hydration; live updates arrive on the mux `unread` channel.
The caller's custom chat-sidebar sections by position, plus the order every movable section (custom, Channels, Direct) renders in.
Move a conversation into one of your sections. Set sectionId to null to return it to its default place. Only your sidebar changes, and other conversations stay where they are.
Create a custom chat-sidebar section for the caller.
Reorder the caller's sidebar sections. Pass every movable section — custom ids plus "channels" and "direct" — top to bottom. Starred stays first and More last.
Rename a chat-sidebar section or set its channels.
Delete a chat-sidebar section owned by the caller.
The caller's chat drafts (inflight + saved), newest first.
Create a saved draft for the caller with text, files, or both. File metadata and suggested thread placement are private references, never access grants. An optional idempotency_key retains one creation identity and binds the ordered file metadata and placement; id requires that key. Matching retries return only { id, replayed: true }, including after discard. Read the draft normally for its current content. Optional source_draft_id and source_revision atomically consume an unchanged caller-owned saved original on the first creation; a newer original stays. A missing or inaccessible source returns 404 without creating a row or receipt. Replay never checks or consumes the source again.
Update a draft owned by the caller. Omitted attachments or suggested_thread_parent_id preserve existing authoring state; [] removes files and null clears suggested placement. This updates the row in place; use DELETE to discard it. Optional if_revision compares the saved Read draftRevision under a row lock; stale input returns 412 without changing the row.
Delete a draft owned by the caller. Optional if_revision compares its saved Read draftRevision atomically; a changed draft returns 412 and remains saved.
Get online status (true = online within the last 5 minutes) for a set of users in the current workspace. Returns user_id → boolean. Capped at 100 user IDs. Fallback/cold-hydrate surface only — the primary presence path is the full-state push over the `presence` /ws/mux channel; poll this on connect or when the socket is down, not on a steady timer.
Exchange JWT/ak_ for a short-lived socket capability for caller workspace/session. No inputs: credential sets identity. Object channels require resource capabilities; API rechecks writes. Principal-rate or instance DB-admission exhaustion: 429 Retry-After.
Exchange credential for a short-lived capability on one requested object channel. REST object role decides authority; operations only narrows it. Unauthorized operations are refused, never clamped. Unreachable/absent object: 404. API rechecks writes.
Cross-module command-palette search — every module including chat messages. `q` is the query string; `limit` caps results (default 30, max 50). Empty query returns an empty response. Module diversity is preserved. `counts` carries the per-module match total before the per-module cap.
Search across every module in the user's workspace (docs, sheets, slides, canvas, tasks, contacts, deals, channels, chat messages, wiki, forms, calendar, users, mail, files, teams, and saved Analytics reports for callers who can read them). Returns a flat list of results with module-aware metadata, plus `counts` — the per-module match total before the per-module cap. Optional module filter narrows the search to one or more modules; `sort` picks relevance or recency. Use this when the user asks to find something but you don't know which module it lives in.
Look up a hover-card style preview of any referenced entity (document, task, contact, deal, channel, chat message, user, event, file, wiki page, form, email). Module keys include `docs`, `tasks`, `channels`, `messages`, `users`. Returns rich details — title, status, owner, due date, amount, etc. — without fetching the full record. Use this when an entity is referenced by ID and you need a quick summary.
List recent workspace activity (last 30 days). Scoped to the current user's workspace. Filterable by resource type, resource ID, and acting user. File activity uses current file view authority for data and totals. A file cohort above the activityFileCandidateMax platform limit returns 503 activity_scope_too_large; narrow by resource_id or user_id.
Compare two repository refs before opening a pull request.
Read the diff for one commit in this pull request.
Compare the caller's last reviewed head with the current head. Historical or truncated source is explicitly reported.
List workspace users and teams available for an explicit repository grant. Repository owner only.
List repositories offered by the workspace GitHub installation. Workspace administrators only.
List repositories the caller can read, with their Code role.
Connect a repository from this workspace's existing GitHub installation. Workspace admin only; grants the connector repository ownership.
Update the default target branch, review policy or connection. A null target resets to GitHub's default branch. Repository owner only.
Read repository access grants. Repository owner only.
List currently eligible Ambiguous reviewers, including agents.
Read the repository labels and milestones available for pull requests.
Permission-filtered pull request inbox, newest update first. Each item is a summary of what an inbox row shows; read one pull request with code_pull_get for its files, commits, description, comment bodies and activity. Pass the returned nextCursor as cursor for the next page; nextCursor is null after the last page. A page may hold fewer than limit items while nextCursor is not null.
Create/adopt a GitHub PR with accountable Ambiguous author. Base defaults to repo target; retry same idempotency key. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Read a native review, its revision, discussion, checks and merge blockers. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Reconcile with current GitHub state. Never interprets an older webhook as current source truth. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Edit metadata, assign reviewers, follow, close, reopen or mark ready. Requires the last read version. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Submit a review for the specified comparison as an authenticated user or agent with repository review permission. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Dismiss a blocking review with an auditable reason. Maintainer only. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Resolve or reopen a review without changing its verdict. The PR author or maintainer may resolve another reviewer's feedback. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Start your draft review, or return the one you already have. Add pending inline comments with create_resource_comment on the draft's id; submitting the review publishes them with its decision. The response carries every file patch, thread and activity entry. Pass `fields` to return only what you need, for example `id,version,blockers,data(mergeIntent,syncError,provider(state,headSha,baseSha,baseRef,mergeable,checks))`.
Export the complete review record with original anchors and attribution as versioned JSON or Markdown.
Describe what happened; only description required. No category or goal outcome; Ambiguous triages. Optional title, priority, context, screenshotDataUrl and attachments. Unauthenticated REST accepted; valid credentials identify reporter. Rate-limited per authenticated user or anonymous IP. Stored as a task in Ambiguous Feedback project; 200 returns task-id receipt. Storage failure 500: retry, no false receipt. Unconfigured intake 503 with {error:"Feedback reporting not available"} (raw response outside contract enum).
Read back one submission by its receipt. Answers only the submitter — anyone else, including other members of the submitter's workspace, gets 404. Returns the public projection: a status word (`filed` | `accepted` | `declined` | `shipped`, derived from the task Ambiguous works), the title, the priority, the task key, and timestamps. Nothing else on the task is exposed.
List the caller's own submissions, newest first, as public projections (see `get_feedback`). Paginated with `limit` (default 50) and `offset`.
List every live API key acting as the caller or as an agent the caller manages, classified by how it connects: `connector` (a chat app approved through OAuth), `terminal` (a terminal tool set up from a setup message) or `api_key`. An admin or owner additionally sees every agent's keys; another member's own keys never appear. `setting_up` is true while a terminal tool has not yet redeemed its setup message.
Read an agent's email scope, approved contacts, record version, the human who reviews its held emails, and whether the calling principal may change it. Readable by workspace humans with standing access, existing management readers, and the agent itself; `can_manage` is false for the agent's own read.
Set an agent's email scope. A human manager or human workspace admin/owner only — an agent cannot widen its own delegation even holding the admin role, and no profile or config update can write this field.
List teams the current user belongs to, with their role (admin or member) in each. Use this when the user asks 'what teams am I in?' or 'my teams' or 'my memberships'.
Search teams visible to the caller for the share dialog. Returns the union of public teams and private teams where the caller is a member. Supports case-insensitive name filtering via `q` and a configurable `limit`.
List workspace teams visible to the current user. Returns public teams and private teams the user is a member of, with member counts. Filterable by `q` (case-insensitive substring match on name + slug). Optional `limit` caps the result count. Use this when the user asks to see teams, distribution lists, or workspace teams.
Create a new team in the workspace. The caller becomes the team admin. Teams can serve as email distribution lists (when email_local_part is set) and/or permission targets. Use this when the user asks to create a team, distribution list, or mailing list.
Paginated member list. Private teams visible only to members or admins.
Add a user to a team with an optional role (admin or member). Only team admins or workspace admins can add members. Default role is 'member'. Use this when the user asks to add someone to a team or distribution list.
Change a member's role. Admin only. Cannot demote the last admin.
Remove a member. Admin only. Cannot remove the last admin.
Get the caller's email-delivery preference for the team. Returns `{team_id, user_id, email_delivery}`.
Update the caller's email-delivery preference for the team. Body: `{email_delivery: 'each' | 'daily_digest' | 'weekly_digest' | 'none'}`. `none` is refused on the workspace-wide all-members list, which has no opt-out.
Join a team. Open teams grant instant membership; request teams create a pending join request; invite teams reject.
Leave a team. Cannot leave if you're the last admin.
Cancel your own pending join request for this team.
List pending join requests. Admin only.
Approve a join request — adds the user as a member.
Deny a join request.
Get details of a single team including its first 10 members. Returns team metadata (name, slug, visibility, join policy, email) and an embedded members list with each member's user_id, display_name, email, role, and added_at. Use this when the user asks about a specific team, its members, or 'who is in team X?'.
Update a team's settings (name, slug, description, email, parent, visibility, join policy, posting policy, icon, color). Only team admins or workspace admins can update. If changing parent, validates no cycle in hierarchy. Use this when the user asks to rename a team, move a team under a different parent, or change team settings.
Permanently delete a team from the workspace. Only team admins or workspace admins can delete. All memberships are removed. Child teams are orphaned (parent set to null). Use this when the user asks to delete, remove, or dissolve a team.
Create a saved insights report.
Update a saved insights report.
Create a saved funnel report.
Update a saved funnel report.
Create a saved retention report.
Update a saved retention report.
Create a saved cohort definition.
Update a saved cohort definition.
Create a new analytics dashboard.
Update an analytics dashboard's name, dashboard query or tile layout; omitted fields are kept. Requires editor on the dashboard's project: a viewer gets 403, and a caller with no role gets 404. Every report tile must come from the same project (`400 project_mismatch`).
Unified live debugger — interleaved ingest successes and errors, cursor-paginated, tenant-isolated via writeKey ownership.
List catalog metadata overrides for the workspace.
List all writeKeys for the workspace (plaintext never returned on list).
Execute an ad-hoc analytics query (events/insights/funnel/retention/cohort). Returns structured results over the project's view (the workspace default project when project_id is omitted). An `events` page can hold fewer events than `limit`, or none, while `hasMore` is true; keep requesting with the returned cursor until `hasMore` is false.
List source-owned Analytics evidence with bounded filters and stable pagination. Measured group IDs never grant access. A page can hold fewer events than `limit`, or none, while `hasMore` is true: a page stops after examining a bounded number of events the project view may not show. Keep requesting with the returned cursor until `hasMore` is false.
Read one Analytics evidence event owned by the caller's source workspace. Foreign and missing IDs return 404.
List saved insights reports in the active workspace, including their names and query definitions. Lists one Analytics project (`project_id`) and requires a role on it.
List saved funnel reports in the active workspace, including their names and conversion query definitions. Lists one Analytics project (`project_id`) and requires a role on it.
List saved retention reports in the active workspace, including their names and query definitions. Lists one Analytics project (`project_id`) and requires a role on it.
List saved cohort reports in the active workspace, including their definitions and stored member counts. Lists one Analytics project (`project_id`) and requires a role on it.
List the saved analytics dashboards of one Analytics project (`project_id`), including their tile layouts. Everyone who can view the project sees every dashboard in it. Requires a role on the report's project; `404` otherwise.
Read a saved insights report's name and stored query definition by ID, without executing the query. Use analytics_insights_list to find report IDs. Requires a role on the report's project; `404` otherwise.
Read a saved funnel report's name and stored conversion query definition by ID, without running the funnel. Use analytics_funnels_list to find report IDs. Requires a role on the report's project; `404` otherwise.
Read a saved retention report's name and stored query definition by ID, without computing retention. Use analytics_retentions_list to find report IDs. Requires a role on the report's project; `404` otherwise.
Read a saved cohort's name, stored criteria and member-count snapshot by ID, without recomputing its audience. Use analytics_cohorts_list to find cohort IDs. Requires a role on the report's project; `404` otherwise.
Read a saved analytics dashboard's name, stored query and tile layout by ID. Requires a role on the dashboard's project; unavailable dashboards return not found. Use analytics_dashboards_list to find accessible IDs. Requires a role on the report's project; `404` otherwise.
Get the workspace's analytics settings — the timezone calendar windows such as today-7d and day/week/month buckets resolve in.
Read one number from a saved insight or funnel over a window: the value, the same metric over the window before it, a sparkline and a link to the report. Embedded metrics on canvas, in chips and in sheets read through this. Example: report=insights:<id>&from=today-28d&to=now.
Per-event last_seen_at and 30-day volume for the workspace's tracked events.
Upsert external metric rows (ad spend, cloud spend, incident minutes) keyed by metric, grain, period and dimensions. A row replaces the live value only when its observed_at is newer; an older observation is kept in the restatement history and reported as stale. Requires workspace admin rank and the analytics.write scope; the public write key is refused.
List metric definitions visible in the named Analytics project, or its workspace default when omitted. management=true instead reads the workspace-wide admin catalog. Each definition includes its newest stored period.
List the Analytics projects the caller holds a role on, each with its view and the caller's role. Workspace admins hold no implicit role: a private project an admin was not added to is not listed.
Read one Analytics project, its view and the caller's role. `404` when the caller holds no role on it.
List the project's members: users and teams with their roles. The Everyone team row is the workspace tier.
Ask the project's owners for access at viewer or editor. Must ask higher than the role held (`400 access_role_not_higher`); `409 access_request_open` while one is open; `404` for an unknown project.
Return whether the caller has an open access request for the project.
Approve at viewer or editor, or decline, an open access request. Project owners only. Returns 409 once decided.
List the workspace's restricted event names, trait keys and metric keys. Workspace admin only.
List a project's annotations, oldest date first: each is a date, an optional end date, a short text and its author, drawn on every time-series chart in the project. `from`/`to` keep only annotations overlapping that range.
Everything shared with the calling account through a direct grant, a team grant, or a bound external grant, across docs / sheets / slides, Drive, Wiki, Forms and Calendar — and across every workspace and identity the account owns. Excludes what the account owns, and excludes workspace-tier access, which is standing authority rather than a share. An API key sees only resource families within its read scopes. Newest share first. `scope=workspace` narrows discovery to resources in the authenticated current workspace without changing access.
Hide one row from the caller's own Shared-with-me list. Revokes nothing — the grant is untouched and the resource stays reachable by its id or link; only the list entry goes. 404 when the caller holds no live grant or the key lacks that resource's read scope. Idempotent.
Undo a "Remove me": restore the exact grant the caller removed, from the `undo_token` the leave call returned. The token is bound to the caller, single-use, and lasts ten minutes. An outside grant is re-checked against the workspace's current outside-sharing policy. 400 `undo_token_invalid` / `undo_token_expired` / `undo_token_used`; 403 `outside_sharing_off` / `outside_sharing_domain_not_allowed`; 404 when the resource is gone or the key lacks its read scope.
Set a flat map of fields on the node output. Native equivalent of the n8n `Set` node — alias `n8n-nodes-base.set → formatter_set`. Pure passthrough: `parameters.values` is resolved upstream by the dispatcher's `resolveActionParameters` walker, so JSONLogic rules referencing `{var: "<UpstreamNode>.<field>"}` produce the literal value the handler echoes back. Downstream nodes reference the result via `{var: "<NodeName>.values.<key>"}`. Spec line 41: formatters are `formatter_*` operationIds.
List reusable signing templates.
Create a reusable signing template from field layout and signer roles. A Doc source is rendered to an independent frozen PDF snapshot. Choose either source_doc_id or source_pdf_file_id.
Create a new draft document from a template, populating fields with placeholder signer roles.
Create a draft signing document.
List signing documents in the workspace.
Get a signing document with signers and fields.
Get the append-only audit trail for a visible signing document.
Update a draft signing document.
Add a signer or CC recipient. An agent user can be a CC recipient but never a signer; adding one as a signer is refused.
Atomically set the sequential signing order: the full ordered list of the document's signer-role ids. Draft only.
Add a signature field to a signing document.
Reassign, reposition, resize, or change requiredness for a field on a draft signing document. The field id is stable — updates never re-create the field.
Remove a field from a signing document.
Remove a signer or CC recipient from a draft signing document.
Validate signers and render the PDF preview (draft → preview_pending).
Download the Certificate of Completion for a completed document.
Verify the integrity and signature of a completed document.
List signing documents that use this document as a source.
Archive an email or return it to the inbox. archived=true archives; false unarchives. Existing mailbox access rules apply.
Mute or unmute a mail conversation. muted=true moves it out of the inbox and keeps future replies muted; false restores it. mailbox_id selects a shared mailbox.
Assign a shared-mailbox conversation to a member, or clear its assignment with user_id=null. Does not change its folder.
Archive or unarchive a chat channel. Archiving freezes writes for everyone. Requires the same administrative authority as the individual archive and unarchive actions.
Mute or unmute a chat channel for yourself, including DMs, temporarily or indefinitely. Muting silences ordinary message, keyword and subscribed-thread alerts and removes the channel-presence badge contribution; direct mentions and broadcasts still reach you. This changes the channel mute setting, not its notification preference or a delivery-suppression rule. muted=true optionally accepts until or duration_ms; omit both for indefinite mute. muted=false cannot include timing fields.
Archive or unarchive a task. Archive is separate from Trash; archived=false does not restore a trashed task. Existing project and task permissions apply.
Subscribe or unsubscribe yourself from a task's notifications. When subscribed=true, user_id may select another subscriber under the existing subscription rules; unsubscribe applies only to yourself. Subscription does not grant task access.
Add or remove favorites (star/unstar) for documents, spreadsheets, presentations, canvases, Drive items, forms, channels, teams or Wiki pages. Resource type is resolved from ID; existing starring/access rules apply. Tasks-sidebar view/project pins use tasks_favorites_add/remove instead.
Read comments on a document, Drive file, task, wiki page, mail conversation, pull request, or code review. Resource type is resolved from ID; a mail conversation can use its thread key with mailbox_id. Every container returns the same comment model: roots with their replies, the caller's capabilities on each, and unread counts.
Create a comment on a document, Drive file, task, wiki page, mail conversation, pull request, or code review. Replies use parent_id. anchor points at a passage, cell, slide, board point or element, Drive file region, or diff line; each container accepts only its own kinds and refuses others with 400 comment_anchor_invalid, and tasks and mail conversations take no anchor. idempotency_key makes a retry return the original comment. Mail comments are internal notes and cannot be sent as email.
Update a comment by its ID. content edits the text and only its author may; expected_content checks the original text and rejects a stale edit with 409 comment_edit_conflict. resolved resolves or reopens the comment's thread, which the thread's author or anyone with edit access may do. anchor moves a root comment on a canvas board to a board point or an element, by the same rule; a reply refuses it with 400 comment_anchor_on_reply and any other container with 400 comment_anchor_invalid.
Delete a comment by its ID. Its author, or anyone with the container's moderate access, may delete it; deleting a thread's first comment deletes its replies.
Add or remove your emoji reaction on a chat message or a comment. The target is resolved from ID. present=true adds; false removes. Existing parent visibility, workspace custom-emoji rules, broadcast-message identity, and missing-reaction errors apply. A comment returns the updated comment.
Mark a comment thread read or unread for yourself by its root comment ID, on a document, Drive file, task, wiki page, mail conversation, pull request, or code review. read=true clears its unread badge; false keeps it flagged until reopened, even when its only comments are yours. A reply ID answers 404; marking a resolved thread unread answers 409 comment_resolved.
Mark every comment thread on a document, Drive file, task, wiki page, mail conversation, pull request, or code review read for yourself. Resource type is resolved from ID; a mail conversation can use its thread key with mailbox_id. Idempotent; read state is private.
List named/team grants, workspace/link tiers and pending invitations for a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar. id names the resource; grant IDs name recipients. To remove broad access but keep explicit collaborators and their roles, use set_visibility(visibility=restricted). The Everyone entry mirrors workspace access, not a named share. Resource kind is inferred from id.
Granting by email sends the recipient a notification email. Share a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar with exactly one user_id or named team_id (workspace members/teams), or email (anyone). For all current/future workspace humans and agents use set_visibility(visibility=workspace, workspaceRole), not per-person grants or the Everyone entry. message applies only to email. Unsupported roles are rejected; re-sharing cannot lower a live grant.
Change an existing named user/team grant on a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar. id names the resource; exactly one user_id or team_id names the recipient. For the workspace tier use set_visibility and workspaceRole, not the Everyone entry. role and expires_at are set exactly, including shorter expiry.
Remove a named grant, pending invitation or link access on a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar. Give exactly one user_id, team_id, pending_share_id (unaccepted invitation), or link=true (restricts the link tier; unavailable on calendars). To remove broad access while retaining explicit collaborators use set_visibility(visibility=restricted), not Everyone removal.
Set broad access for documents/sheets/slides/canvas/Drive/forms/wiki/calendars, not individual grants or Everyone entries. `restricted` keeps the owner and explicit collaborators and removes broad tiers; `workspace` admits current and future workspace humans and agents at workspaceRole. Result url is the product URL; never share /api/public snapshots. Unsupported fields rejected.
Remove your own direct grant on a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar. The owner cannot leave. Returns an undo_token that shared_with_me_undo_leave accepts for ten minutes.
Ask the owner of a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar, task, project, or CRM pipeline for access at a role higher than any you hold. Documents, Drive items and wiki pages offer viewer, commenter and editor; forms and calendars viewer and editor; projects commenter and editor; tasks editor; pipelines member. Returns 409 while a request is already open.
Whether you have an open access request for a document, spreadsheet, presentation, canvas, Drive item, form, wiki page, or calendar, task, project, or CRM pipeline.
Approve an open access request at a named role, or decline it. Only principals entitled to share the resource may decide. Returns 409 once decided.
Read from the server on 2026-11-18, including each tool's own safety annotations.
*; narrowing happens through your workspace role and API key scopes, not the OAuth screen./mcp serves; counts will differ by URL.llms.txt (retrieved 2026-11-18) · retrieved 2026-11-18initialize, tools/list (914 tools) and resources/list against the endpoint without credentials, and a comparison tools/list against app.ambiguous.ai/mcp (1,020 tools) (2026-11-18) · retrieved 2026-11-18app.ambiguous.ai/.well-known/oauth-protected-resource/mcp/anthropic, scopes_supported: ["*"] (2026-11-18) · retrieved 2026-11-18Connect Ambiguous once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.