It reads the official document and hands back the part that answers the question, with the anchor to check it against. That is a different thing from a model recalling what the specification said when it was trained.
An npm-installed MCP server that wraps the official Model Context Protocol specification: it builds a navigable tree of the document, finds sections by keyword with anchor links, returns single sections on demand, caches what it fetches, and checks the upstream repository for changes.
- A tree of the specification generated to the depth you ask for
- Sections found by keyword, with direct anchor links
- One section returned on its own, rather than the whole document
- A check against the upstream repository for specification changes
- Cached copies, so repeated questions do not refetch
Node and npm — install @lepion/mcp-server-specnavigator globally, or build from source. Nothing to authenticate: it reads the public specification repository.
