Summary
Reading a dependency's docs stops meaning a browser tab and a keyword guess.
Search answers only for packages whose embeddings already exist, so fetch is a real first step rather than an optimisation, and because both tools take a version, a dependency bump means fetching again. What that buys is a question asked the way you would ask a colleague, answered with the passages that bear on it.
What it is
Two tools over Elixir Hex package documentation: one downloads a package's docs and turns them into vector embeddings, the other answers a natural-language query against what has been embedded.
What you get
- Documentation for a named Hex package and version downloaded, converted to markdown, split into semantic chunks and embedded
- Natural-language search across one or more packages, returning the documentation snippets that match
- A version on both tools, so an index belongs to a specific release rather than to the package in general
- A limit on how many snippets come back per query
- A force flag to rebuild embeddings for a package that already has them
Requirements
Node on your PATH. No account and no key — the only prerequisite is that a package has been fetched before you search it.
Setup effort
One command — npx -y hexdocs-mcp@0.5.0