Search and ask the Metabase docs
Find the exact section of the Metabase documentation that answers your question, for the Metabase version you run. Results link to the heading on the docs page, and name the plan a feature needs.
Examples
What is indexed
- The Metabase docs for every supported version (an LTS version and the newest releases), and the Metabase Cloud docs.
- The REST API reference (one entry for each endpoint) and the embedding SDK reference.
- The Learn tutorials, the glossary, and articles from metabase.com.
The index follows the docs on GitHub. When a docs page changes, search shows the change within about 15 minutes. Search results never come from a language model. They are passages of the docs, ranked.
API
The API is free and needs no key. Each client can make 120 searches a minute. Responses allow any origin, so you can call the API from a browser. The OpenAPI description lists every endpoint and response.
Search
curl 'https://docs.metabase.fyi/v1/search?q=row+and+column+security&version=v0.64&limit=5'
The response lists passages in rank order. Each passage has the page url, the
heading anchor, a snippet, the plan it needs (if any),
its version, and a relevance score from 0 to 1.
q- The question or search terms, up to 1,000 characters. Required.
version-
Your Metabase version, for example
v0.64or0.58. The default is the newest version. If your version has no good match, newer docs are searched too, and the response says so infallback. limit- The number of passages, from 1 to 30. The default is 8.
Answers
POST /v1/ask answers a question from the docs, and cites each claim as
[n] with a link to the docs section it comes from. If the docs do not answer the
question, the answer says so. Send JSON:
{"question": "…", "history": […], "version": "v0.64", "context": {…}}. The answer
streams as server-sent events (retrieval with the sources, then token,
then done), or comes as one JSON object if you send
Accept: application/json. Each client can ask 10 questions a minute.
AI agents
An MCP server at https://docs.metabase.fyi/mcp gives AI coding tools and agents the
same search, the full text of each docs page, and cited answers. It uses Streamable HTTP and
needs no key. For example, in Claude Code:
claude mcp add --transport http metabase-docs https://docs.metabase.fyi/mcp
search_docs- Ranked passages of the docs, each with a link to its section, its version and the plan it needs.
read_doc- A whole docs page as Markdown, by URL or by version and path.
ask_docs- A short answer with numbered citations. Each client can ask 10 questions a minute.