MCP Server
Documentation tools
Seven tools, available as soon as the server starts. They need no credentials and read published content only, so nothing unpublished leaks to an agent.
The tools
| Tool | What it does |
|---|---|
search_docs | Search the documentation for an endpoint or concept |
get_endpoint | Full documentation for one endpoint |
list_endpoints | List endpoints, optionally filtered by section |
get_changelog | Recent API changes, including breaking ones |
list_sdks | Available client libraries, versions, and download links |
get_guide | The full content of a guide by slug |
get_sdk_methods | Ready-to-use SDK snippets per endpoint, in every language you generate |
How an agent uses them
A typical integration task chains several:
- search_docs
The agent looks for the concept it needs: "create a payment".
- get_endpoint
It pulls the full parameter list and response shape for the endpoint it found.
- get_sdk_methods
It fetches the real snippet for your SDK in the user's language, rather than inventing a call signature.
That last step is the one that changes the output. An agent guessing at your SDK's method names writes code that compiles in its head and fails on your machine.
get_sdk_methods
Snippets come from your generated SDKs, so they reflect your actual client shape: the namespacing you chose, the argument style, the method names you overrode.
The agent writes what this tool returns. A method named postV1UsersUserIdContact because the spec lacked an operationId is what lands in your user's codebase.
Only published content
get_guide serves published guides only. A guide you can see in the studio is not one the agent can read, which is correct: unpublished means unpublished, for agents as well as readers.
get_changelog
Returns recent API changes including breaking ones, which lets an agent answer "what do I need to change to upgrade?" against your real history rather than guessing from version numbers.