Within
by Within
Query your organization's process index, observations and source documents from an AI agent. 17 read-only tools, matching Within's own published tool reference exactly. OAuth or API key, and no scope boundary at consent.
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.
Connect Within via MCP
https://api.within.ai/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Within Tools & Capabilities (17)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- No scope boundary exists at consent. Within's RFC 9728 resource descriptor declares
scopes_supportedas an empty array, and its authorization server advertises four scopes —openid,profile,emailandoffline_access. Those identify the user and refresh the token; none of them separates one part of the workspace from another. The empty array matters less here than it would on a write-capable server, because the surface it fails to partition is read-only, but a reader should not mistake the sign-in screen for a permission choice. - Read-only is current, not guaranteed. Within's limitations page states the tool catalog grows over time and names write tools for safe operations as coming. Verify against the tool reference rather than against this page if the distinction matters to you.
- One workspace per connection. Within scopes each connection to a single workspace, switched by reconnecting on OAuth or by generating a key in the other workspace. An agent cannot read across two workspaces in one session.
get_observation_citationonly works for your own sessions. The most detailed evidence tool is restricted to the authenticated user's captures, so a manager cannot pull a colleague's activity timeline through it.- Cross-observation analysis is not a tool. Within states that the pattern most teams want — the same workaround appearing across several captures — emerges only by aggregating the observation set, that Within Advisor does this natively, and that over MCP you stitch it yourself by pulling citations and diffing against the pinned process version.
- Search snippets are not answers. Within's operating principles state that
search,search_processesandsearch_artifactsall return short snippets meant to locate an object, and that an assistant must chain into agetorfetchcall before answering. A page or agent quoting a snippet as a finding is misusing the tool by the vendor's own account. - There is no cross-document text search. Within states plainly that
search_artifact_textworks within one artifact and that no cross-document text search exists on the external surface. - We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail or machine-readable safety hint is published here. Parameter names cited above come from Within's documentation.
- Bulk edits, workspace configuration and user management stay in the UI. Within names all three as out of scope for the MCP.
- It requires a commercial Within workspace. No free tier is documented.
Frequently asked questions
No. Within states that every tool in production is query-only and that the MCP cannot create or modify processes, artifacts, observations or graph entities. All edits happen in the Within UI. This is one of the rare connectors where the vendor's read-only claim survives a name-by-name check of the whole tool list.
None that narrow anything. Within's resource descriptor declares scopes_supported as an empty array, and its authorization server advertises only openid, profile, email and offline_access. Those are identity claims and token refresh, not permissions over your workspace. Consent grants the whole read surface at once.
Seventeen, and Within publishes all seventeen itself. Anthropic's directory and Within's own tool reference name exactly the same set with no drift in either direction. Within adds that the catalog grows over time and names write tools for safe operations as a direction of travel, so treat seventeen as current rather than permanent.
Almost always because the MCP has not been switched on for your workspace yet. Within documents a one-time setup in Settings then MCP, where an admin picks the workspace to expose, and warns that both OAuth and Bearer-token requests fail until it is done. Your organization may also need to approve the connector.
Colleague activity, named and timestamped. Anthropic's directory entry states the connector can see your name and email and list recent workspace sessions including who created them. Within's activity timelines record what a user did, when, and on which systems. Within restricts get_observation_citation to your own sessions.
Not by itself, but its documentation is built around chaining that does. Within publishes a guide pairing its read tools with Gmail, Slack, Linear, Notion, Calendar and Drive connectors, with example prompts that draft mail and post channel summaries. The outbound capability belongs to the second connector; Within supplies the content.
At the auth layer, not inside the assistant. Within scopes each connection to one workspace. On OAuth you disconnect the Within connection in your client and reconnect, and the reconnect flow shows a workspace picker. API keys are bound to the workspace that generated them, so you create a new key there and swap the Bearer token.
No. Within states plainly that it is a commercial service requiring an active workspace and a provisioned user account, and that the MCP is available to enterprise customers. There is no free or trial tier documented for the connector. Organizations without Within are directed to contact the vendor at hello@within.ai.
Sources
- Within MCP developer documentation, machine-readable full text — and index (retrieved 2026-08-22). Within's
robots.txtcarriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— synthesis is expressly permitted by the publisher. · retrieved 2026-08-22 - Within tool reference (retrieved 2026-08-22), with the per-group pages for entry points, process index, observations, artifacts, workspace and Agent Builder · retrieved 2026-08-22
- Within security and architecture (retrieved 2026-08-22) · retrieved 2026-08-22
- Within limitations (retrieved 2026-08-22) · retrieved 2026-08-22
- Within authentication (retrieved 2026-08-22) · retrieved 2026-08-22
- Within FAQ (retrieved 2026-08-22) · retrieved 2026-08-22
- Within guide, "Chain with other connectors" (retrieved 2026-08-22) · retrieved 2026-08-22
- Within Claude and Claude Code install pages — · (retrieved 2026-08-22) · retrieved 2026-08-22
- Within Claude Code plugin repository (retrieved 2026-08-22) · retrieved 2026-08-22
- Live OAuth posture check: anonymous
initializereturned 401 with a www-authenticate challenge naming the resource metadata document, RFC 9728 metadata athttps://api.within.ai/.well-known/oauth-protected-resource/mcp, and authorization server metadata athttps://api.within.ai/.well-known/oauth-authorization-server(2026-08-22) · retrieved 2026-08-22 - Anthropic Connectors Directory entry — , read from our directory snapshot dated 2026-08-16 · retrieved 2026-08-16
- Within support — <mailto:hello@within.ai> · Privacy
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Productivity
- Developer
- Within
- Tools
- 17
- Domain
- api.within.ai
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.