Summary
Four ways in, graded by how much of the documentation the answer actually needs.
The grading is the design. The index orients, search finds a keyword, get_docs_page answers a known topic, and get_docs_full puts the whole set in context — its own description warns that this returns a large volume of text and points you back at the narrower tools. That matters because the two failure modes of documentation retrieval are opposite: pulling too little to answer, or so much that the answer is buried in what came with it.
What it is
A documentation-retrieval server holding libraries and frameworks as complete sets, each addressed by the docName that list_docs returns.
What you get
- list_docs, returning every available documentation set with its name, description and source URL — and the docName the other four tools require
- search_docs, searching inside one named set and returning matching pages ranked by relevance with their paths and descriptions
- get_docs_index, a condensed index of a set's pages for working out what is covered before you ask anything
- get_docs_page, the full content of one page by its relative path
- get_docs_full, the entire documentation set consolidated into a single response
Requirements
Nothing — no account, no key. Every tool except list_docs takes a docName, so list_docs is the required first call.
Setup effort
One command — npx -y @cartographai/atlas-docs-mcp