Context7 vs ref-tools-mcp
Context7: Context7 is an MCP server that injects up-to-date, version-specific library documentation directly into LLM prompts. Add "use context7" to any coding prompt and it fetches current docs for the library you're working with, eliminating hallucinated APIs and outdated code examples. Works with Claude Desktop, Cursor, Windsurf, and any MCP-compatible editor.; ref-tools-mcp: Ref MCP is a server that enables AI coding tools and agents to access technical documentation efficiently. It optimizes token usage by filtering search results and extracting only the most relevant parts of documentation pages, helping models stay focused and reducing operational costs.
Context7 resolves a library and returns its docs and examples. Ref focuses on trimming what comes back to the most relevant part.
TL;DR
- You use Cursor, Claude Code or another coding agent with fast-moving public libraries and want current, version-specific docs and examples.
- You want the least setup: `npx ctx7 setup` signs you in and configures the skill or MCP server for your agents.
- You want to work with or without MCP: the CLI plus skill mode fetches docs without an MCP server.
- You need docs for internal or private libraries: a self-hosted index fits better.
- Your data policy rules out sending queries to an external service: requests go to Upstash's hosted backend.
- You need a fully open-source stack: only the MCP server is open source; the API backend, parser and crawler are private.
- Context size is your main concern: page reads return about the most relevant 5k tokens, and repeat searches do not return the same results in a session.
- You want two simple tools, search and read-URL, in Claude Code, Cursor or another MCP client.
- You want to search private resources such as repos and PDFs alongside public docs.
- You cannot send queries to a vendor: it needs an API key and the hosted ref.tools service.
- You want the recommended HTTP server: this repository is the legacy stdio server.
- You need independent proof of the token savings: the README's numbers are the project's own example traces.
Side-by-Side Comparison
Features
Use Cases
Best For
FAQ
Deploy on DigitalOcean
Host your MCP servers, Docker containers, and AI agents with $200 free credit over 60 days.