Summary
Version-correct documentation arrives without the browser detour.
The split between the two tools is the whole design: search_docs returns sections with their URLs so the model can choose, and fetch_doc then pulls that exact page as markdown rather than a summary of it. Because the second call takes any URL from the first, trimming an anchor or dropping to a parent path turns the same tool into a table of contents — useful when the question is what a library covers rather than how one function behaves.
What it is
A two-call documentation lookup: search a library's indexed docs, then fetch a page from the results in full.
What you get
- Search over a library's indexed documentation, returning sections with titles, summaries and URLs, drawn from official versioned docs
- A library addressed either by simple name or by an exact owner/repo pair when the simple name is ambiguous
- Any result URL retrieved in full as rendered markdown, with code examples and API signatures intact
- A URL trimmed to its parent path to get a broader table of contents with section previews instead of one page
- A tokens parameter on the search, so how much documentation you pay context for is set rather than accepted
Requirements
A Node runtime and network access to the documentation index. No account and no key.
Setup effort
One command plus a key — npx -y docfork, then supply credentials