Gives Claude direct access to technical documentation through two core operations: search and read. The search tool takes natural language queries and returns relevant docs from public sources or private repos, while the read tool fetches and converts URLs to markdown. The implementation is designed around token efficiency, it filters out already-seen results in repeated searches and intelligently truncates long documentation pages to return only the most relevant 5k tokens based on your search history. Works over streamable HTTP or stdio. Useful when you need your agent to pull context from API docs, libraries, or internal wikis without burning through tokens on irrelevant content or dealing with context rot from oversized documentation dumps.
claude mcp add --transport stdio tools.ref-ref-tools-mcp uvx ref-tools-mcp