Appearance
Connect an MCP client
An MCP client is the application that connects to TKP and calls its tools. It may be a chat app, a coding assistant or your own application. Choose your client below; for a framework-free Python example, use the developer quickstart.
Before connecting, find your TKP application address and choose how the client will sign in:
- Personal Access Token: create a token with the MCP interface enabled and read permission for a library. The client sends it as a bearer token in an Authorization header.
- OAuth: use a client that supports browser sign-in. TKP asks which workspace and libraries it may access during authorization.
In every example, replace {APP_URL} with the application address, such as https://tkp.example.com, and <token> with your token. Do not use the docs site's address.
| Client | Setup |
|---|---|
| Goose | Remote extension |
| Claude Code | HTTP server with a token or OAuth |
| Claude Desktop | Remote connector or local bridge |
| OpenAI | Responses API and ChatGPT |
| LangGraph | Load tools in Python |
After connecting, ask the client to list the libraries available to it. Confirm that it returns the library you authorized before trying search. To exercise search, that library must contain a file whose processing has finished.
Start with read permissions. Enable writes or expensive analysis only when your application has a way to obtain the user's approval for them. Client-side approval settings do not replace TKP's library permissions.