Read one tool
Includes the input schema and the side-effect class.
Authorization
memberToken The tools internkim uses inside a company, callable from outside it. A token belongs to a person, and every call runs as that person, so a token can never do more than its owner can.
The first token is made signed in, from the settings screen. After that a token makes its own
successors: POST /token for a new one, DELETE /token for the old one, so a script rotates
without a browser. A token can neither mint a rung above its own nor revoke itself.
Send Authorization: Bearer <token> on every request.
In: header
Path Parameters
Response Body
application/json
text/plain
text/plain
curl -X GET "https://example.com/tools/string"{ "name": "string", "namespace": "string", "description": "string", "version": "string", "sideEffectClass": "string", "requiresApproval": true, "requiresUserPresence": true, "inputSchema": {}, "outputSchema": {}}Look at what a call would touch POST
For a tool whose call names a record with a hint, ask what that call would touch. `input` is the same input the call takes and nothing is written. The hint resolves through the same function the call itself uses, so the row previewed and the row touched cannot differ. This is for a client that puts an irreversible action to a person first: narrow the input with the answer's `target.inputField` and `target.id` and the approved call will not resolve the hint again. A tool with nothing to look at answers `target: null`.
document_read POST
Read a workspace document and return Markdown content. path takes an exact /workspace path, or an attachment's exact url copied verbatim from the conversation — a document attached to an earlier message is fetched by that url. Never invent a filesystem path from a url. Use image_read for image files. `read` · v1