Skip to main content
Tranthor runs one remote Model Context Protocol server. Claude, ChatGPT and other MCP clients use it to work in a Tranthor workspace. To connect as a person, follow Connect Claude or ChatGPT. This page is the reference for developers.
The server returns text and structured data only. It has no app screens or embedded views.

Server

Every call acts as the person who connected, within their role (see Roles). Nothing a tool takes from the conversation can change the workspace it works in.

Sign-in

A request without a valid token gets 401 and a WWW-Authenticate: Bearer resource_metadata="..." header that points at the protected resource metadata.
  • Client registration. Client ID Metadata Documents (CIMD) are preferred. Dynamic client registration (RFC 7591) is the fallback, at the registration_endpoint in the authorization server metadata.
  • PKCE. Every authorization uses S256.
  • Resource. Send resource=https://api.tranthor.com/mcp with the authorization request. Access tokens are bound to that resource. A token without it, or one issued for a different resource, is refused with 401.
  • Refresh. Ask for the offline_access scope to receive a refresh token, so the person stays connected without signing in again.

Scopes

The person chooses what to allow on the consent screen. A scope is a ceiling, not a grant: each call also re-checks the connection, the person’s membership and their role.

Roles

A tool the connection can’t use is not listed, so a Viewer’s assistant never sees the draft tools.

Tools

Every tool has a title and annotations. Read tools set readOnlyHint: true. Every other tool sets destructiveHint: true, so a client can ask before it runs. All tool names are 27 characters or fewer.

Read tools

Need tranthor:read. Open to every role. Rates are fractions from 0 to 1 in most tools. In query_analytics, campaign open, click and conversion rates are percentages from 0 to 100.

Tools that save or change work

Need tranthor:write. A draft does nothing until someone sends, starts or publishes it. No tool publishes, approves or deletes.

Tools that send

Need tranthor:send, which the person ticks separately when connecting. Minimum role: Workspace manager. A send follows the preview it was bound to. A preview is good for one send and expires after a short time, so a send that follows a changed message or audience needs a new preview. A send that reaches the whole audience, more customers than the person has approved before, or a new channel or sender is Critical: it is approved inside Tranthor only, and the tools answer with a link instead.

Endpoint for Google

Gemini and Google Search use a separate, fixed endpoint, https://api.tranthor.com/mcp/google/v1, because Google freezes a published app’s tools. It has its own OAuth resource, and a token for one endpoint is refused on the other. It offers 16 tools: the read tools for segments, journeys, broadcasts, message templates, results, analytics and Needs you, plus the four draft tools. It never sends, starts, pauses or approves anything, and its tokens never carry tranthor:send. A change to its tools goes on a new version of the endpoint, and the old one stays up until Google retires it.

Errors

Tools return a message that says what to fix: an unknown id, a role that does not allow the tool, a send that needs a new preview, or an item that is not in the connected workspace. A malformed id is rejected before any lookup.

Activity and revoking

Every write is recorded in the connection’s activity log, and items it creates are marked with who made them and through which assistant. The person can see the log and revoke the connection in Settings, AI Agents. Revoking ends access at once.