Idiolect
by Idiolect
Load a personal writing-voice profile into an AI assistant, then draft, rewrite and score prose against it. Idiolect documents 10 tools; Anthropic's directory lists 4, and the 6 it omits include every tool that stores your writing. OAuth 2.1, no application scopes.
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 Idiolect via MCP
https://idiolect.app/api/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Idiolect Tools & Capabilities (10)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Anthropic's directory under-lists this connector by six tools, and the omitted six contain all four writes. Audit the vendor's connector documentation, not the directory entry, before deciding what this server can do.
- Idiolect's own enumerations disagree. Four sources give four, six, six and ten. The published local-server source gives five and two depending on mode, against documentation describing that package as a six-tool surface. Every count on this page names the source it came from.
- No OAuth scope of any kind exists. Neither the resource descriptor nor the authorization server declares
scopes_supported, so connecting grants whatever the server implements, now and as it grows. There is no read-only credential. - We could not read the remote server's tool schemas or safety annotations. The endpoint returned 401 to an anonymous request. Parameter detail on this page comes from Idiolect's connector documentation and annotations come from its published local-server source — a different deployment, and not evidence about the remote one.
- A copy of your writing is retained for up to 12 months. Pseudonymized, opt-out in Settings, partially human-reviewed, and explicitly covering writing sent from connected AI assistants. The directory description's "Your writing samples stay private" is true about public exposure and does not describe this retention.
- Your prose reaches third-party processors in the United States, including OpenAI for generation and Modal for scoring. Do not route confidential material through the connector on the assumption that it stays with Idiolect.
- The free beta is a switch, not a design. Idiolect documents a complete dormant credits contract with
purchase_requiredandaccounting_unavailableerror states and says enforcement is disabled. The generation tools carry no idempotency key, unlikecreate_my_voice. get_credit_statusis absent from the directory listing, so an agent configured from it has generation tools and no accounting tool. This is harmless only while the unit cost is zero.- Rate limits are 30 calls per five minutes for writing and scoring.
- Voice Match is not a detector and not authorship proof. Idiolect states this on its methodology page and in its terms. Short fragments, genre shift and too few samples distort the score, and it is withheld entirely below roughly 600 characters across four pieces.
- The directory description mentions a Pro tier that the current docs do not. Anthropic's entry says "Pro unlocks full ghostwriting"; Idiolect's connector documentation says all 10 tools are free during beta, and its terms refer only to "legacy" Pro subscriptions. Treat the directory copy as stale.
- The connector's usefulness depends on the model honouring an instruction. Idiolect's server ships MCP
instructionstelling the assistant to load your voice before writing. That is a request, not an enforcement, and no tool verifies the model complied.
Frequently asked questions
Ten, according to Idiolect's own connector documentation, which states plainly that all 10 tools are free during beta. Anthropic's directory lists only four. Idiolect publishes three further enumerations that disagree with both, naming four, six and six tools respectively, so the count you see depends entirely on which page you read.
Six, and they are the consequential ones. The directory omits create_my_voice, approve_voice_evidence, record_voice_feedback, select_onboarding_method, get_idiolect_state and get_credit_status. Every tool that writes something durable to your Idiolect account is in that omitted set, while all four listed tools read a profile or process text in place.
No. Every tool returns text to the calling assistant and nothing else. draft_in_voice takes a format parameter accepting values like linkedin, x_post and newsletter, but that shapes the prose only. Source-reading the published server confirms all outbound requests go to Idiolect's own API, with no social, email or webhook destination anywhere in the dispatch path.
None that narrow anything. Idiolect's RFC 9728 resource descriptor declares no scopes_supported field at all, and its Scalekit authorization server metadata declares none either. Consent is therefore all-or-nothing: no grant separates loading your voice card from creating a profile that stores writing you supplied.
Some of it, and for longer than the connector transport suggests. Idiolect's privacy policy keeps a pseudonymized evaluation copy of your writing for up to 12 months, explicitly covering connected AI assistants, and says some of it is reviewed by their team. You can switch this off in Settings, which also removes entries already held.
Not during the free beta. Idiolect's terms state that drafts and rewrites consume neither starter nor purchased credits, and its documented get_credit_status response returns a unitCost of zero with the balance flagged as a preview. That accounting tool is not among the four Anthropic lists, so a directory-configured agent cannot check spend.
There is not enough of your saved writing to score against. Idiolect's troubleshooting states the threshold as roughly 600 characters across four separate pieces, and that it withholds the number rather than showing one that does not mean anything. You still receive the draft; add more samples to your Voice Profile to get the score.
Yes. Idiolect's connector troubleshooting states that writing and scoring are capped at 30 calls every five minutes, and that the limit clears on its own with no action needed. Idiolect's terms separately forbid attempting to disrupt or abuse the Service or its rate limits, so treat the cap as a contractual boundary rather than a soft target.
Yes, through a separate npm package rather than the remote connector. Idiolect documents codex mcp add with the idiolect-mcp package and a personal API key. That local server is a different and smaller surface: its published source registers five tools in API-key mode and two in the older slug mode, not the remote server's ten.
Sources
- Idiolect MCP connector documentation (retrieved 2026-08-23). The complete 10-tool table, input parameters, example responses and troubleshooting. Nominated indirectly: the directory's documentation URL is the
/developersoverview, and this sibling path carries roughly 4x the content. · retrieved 2026-08-23 - Idiolect developer overview (retrieved 2026-08-23). The URL Anthropic's directory nominates; lists six tools. · retrieved 2026-08-23
- Idiolect machine-readable index,
llms.txt(retrieved 2026-08-23) · retrieved 2026-08-23 - Idiolect full answer-engine context,
llms-full.txt(retrieved 2026-08-23). States that/developers/mcpdocuments "all 10 tools". · retrieved 2026-08-23 - Idiolect Voice Match methodology (retrieved 2026-08-23) · retrieved 2026-08-23
- Idiolect canonical answers (retrieved 2026-08-23) · retrieved 2026-08-23
- Idiolect pricing (retrieved 2026-08-23) · retrieved 2026-08-23
- Idiolect Privacy Policy, retention and sub-processors (retrieved 2026-08-23) · retrieved 2026-08-23
- Idiolect Terms of Service, credits and acceptable use (retrieved 2026-08-23) · retrieved 2026-08-23
- Published MCP server source,
Moshpit-Labs/idiolect-mcpat commitf7a2bafc6239d8d4c69064ec282f28b9dcf3d6d5(read 2026-08-23). Tool registrations, annotations, capability block and the full outbound dispatch path. · retrieved 2026-08-23 - Local package
idiolect-mcp0.4.0, read from the npm registry API and its published tarball (retrieved 2026-08-23). Registers four account-mode tools against the repository's five. The npm web page atnpmjs.com/package/idiolect-mcpreturns 403 to a plain client, so the registry API was used instead. · retrieved 2026-08-23 - Official MCP registry entry
app.idiolect/idiolect-ai(retrieved 2026-08-23). Names the current repository; an earlier version pointed at111mihir/idiolect, which now 404s. · retrieved 2026-08-23 - Live auth posture check: anonymous initialize to
https://idiolect.app/api/mcpreturned 401 with awww-authenticateBearer challenge; RFC 9728 metadata athttps://idiolect.app/.well-known/oauth-protected-resource/api/mcp; authorization server metadata athttps://idiolect.scalekit.com/resources/res_128346102637266216/.well-known/oauth-authorization-server(2026-08-23) · retrieved 2026-08-23 - Idiolect
robots.txt(retrieved 2026-08-23). Allows every page cited here by name, including for ClaudeBot and Claude-User, and carries noContent-Signaldirective./api/is disallowed to crawlers; no/api/path was fetched for content. · retrieved 2026-08-23 - Anthropic Connectors Directory entry — , read from our directory snapshot dated 2026-08-16 · retrieved 2026-08-16
- Idiolect support — <mailto:hello@idiolect.app> · 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
- Idiolect
- Tools
- 10
- Domain
- idiolect.app
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.