Ref MCP
An MCP server that gives a coding agent token-efficient access to documentation for public and private APIs, services and libraries, so it can look things up instead of guessing.
Ref MCP is a Model Context Protocol server that gives an AI coding tool or agent access to documentation for APIs, services, and libraries, aimed at keeping an agent up to date without wasting its context window. Its search tools are designed to match how a model searches, returning only what is needed: a simple query typically costs the agent a couple of small tool calls, and the model can then choose a few results to read in more depth.
Within a session, Ref tracks earlier searches so it never returns results the agent has already seen, letting the agent both page through results and adjust its query at the same time, and when reading a documentation page it uses the session's search history to drop less relevant sections and return roughly the most relevant five thousand tokens, instead of the tens of thousands of tokens a large page can otherwise pull into context.