MailcrateDocs for AI Agents
Let a coding agent search these docs over MCP, with or without an API key, or read them as plain markdown through llms.txt.
An agent writing code against Mailcrate does better asking the docs a question than reading a website. Everything here, the API reference included, can be searched and read from inside the agent's own session. This works over MCP, and for agents that browse instead, as plain markdown.
Over MCP
Two tools, on every Mailcrate MCP server:
| Tool | What it does |
|---|---|
search_docs | Keyword search over the guides, the developer docs and the API reference. Returns the best-matching sections, each with a snippet and a path. section narrows it to reference, developers or guides. |
get_doc_page | One page as markdown, or one section of it when the path ends in an #anchor. API reference pages give the method and URL, the scope the key needs, every parameter and body field with its type and constraints, and the response fields. |
An agent typically searches first, then reads the one section that answers the question. For a long page, that costs far less context than reading the whole page.
Without an API key
The docs are public, so searching them needs no workspace. The hosted server serves the two docs tools, and nothing else, at a second endpoint that takes no credential:
claude mcp add mailcrate-docs --transport http https://mcp.mailcrate.io/docsThe CLI does the same over stdio with --docs-only:
claude mcp add mailcrate-docs -- mailcrate mcp --docs-onlyWith an API key
If the agent is already connected to your workspace, through the
hosted MCP server at https://mcp.mailcrate.io/mcp or
mailcrate mcp, it has both docs tools alongside the
workspace tools. No second server is needed. The docs tools never call the API,
so they spend none of your key's rate limit.
From a terminal
The same two tools are CLI commands, and they need no key either:
mailcrate search-docs "verify webhook signature"
mailcrate get-doc-page /developers/webhooks#verify-the-signatureSearching well
Search is by keyword, not by meaning, so a query works best in the words the
docs use: webhook signature, idempotency key, 429 retry. Filler words like
"how do I" are ignored, small typos are forgiven, and webhook also finds
webhooks. If the first query misses, rephrase it. Agents are good at that, and
it works better than a longer sentence.
The search covers the docs as of the Mailcrate version you are connected to:
the hosted server is updated with every deploy, and mailcrate mcp with every
CLI release.
As plain markdown
For agents that fetch URLs instead of calling tools:
| URL | What it is |
|---|---|
https://docs.mailcrate.io/llms.txt | An index of every page, with its description, in the llms.txt format |
https://docs.mailcrate.io/llms-full.txt | Every page in one markdown file |
https://docs.mailcrate.io/llms.mdx/<path>/content.md | One page as markdown. This is what each page's Copy Markdown button copies. |
https://docs.mailcrate.io/openapi.json | The public API's OpenAPI spec |
llms-full.txt is the right choice when an agent should keep the whole picture
in context. For anything narrower, the MCP tools are cheaper.