MCP Reference Servers
Official Model Context Protocol reference servers — the canonical examples for filesystem, git, memory, and more.
About
modelcontextprotocol/servers is the official collection of MCP reference implementations maintained by the MCP steering group at Anthropic. It houses the canonical servers every MCP developer learns from: filesystem (scoped file operations), memory (knowledge-graph persistence), sequential-thinking, git, fetch, time, and the everything test server — as TypeScript (npx) and Python (uvx/pip) packages. The repo also indexes official company integrations and community servers, and the MCP Registry lists published servers for production use.
The honest picture: these are explicitly reference implementations for education, not production-ready solutions — the README says so directly, and legacy servers (GitHub, Slack, SQLite, Brave Search) were moved to a separate archived repo. For production tooling, prefer maintained community or vendor servers from the registry. Still, as the spec's living documentation and the fastest way to prototype an MCP client setup, it's indispensable.
What it can do
- Reference servers: filesystem, memory, git, fetch, time, sequential-thinking
- Published npm and PyPI packages
- Index of official and community integrations
- Educational examples of MCP SDK usage
- MCP Registry for production server discovery
Setup
Run servers directly: npx -y @modelcontextprotocol/server-memory or uvx mcp-server-git. Add to Claude Desktop via the mcpServers JSON config.
Prerequisites
Node.js (npx) or Python (uvx); an MCP-compatible client like Claude Desktop or Claude Code.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 47; Trending score 90/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.