MCP Reference Servers: The Official modelcontextprotocol Repo
The official MCP reference servers repo: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time, plus where archived servers moved.
By Claudekit Team Published 1 min read
- Type
- MCP Server
- Pricing
- Open Source
- Repository
- github.com
- GitHub stars
- 91,529
- Website
- modelcontextprotocol.io
The modelcontextprotocol/servers repository is maintained by the MCP steering group and houses a small set of reference servers. They show how MCP features and the official SDKs work, and several are handy in daily use. If you are new to MCP, start with our MCP servers guide.
The current reference servers
| Server | What it does |
|---|---|
| Everything | Test server exposing prompts, resources and tools |
| Fetch | Fetches web content and converts it for LLM use |
| Filesystem | File operations limited to allowed directories |
| Git | Read, search and manipulate Git repositories |
| Memory | Knowledge-graph-based persistent memory |
| Sequential Thinking | Step-by-step, revisable problem solving |
| Time | Time and timezone conversion |
Many servers that used to live here (GitHub, GitLab, PostgreSQL, Puppeteer, Brave Search, Sentry, Google Drive and others) are now in modelcontextprotocol/servers-archived. For a broader catalog, the README points to the MCP Registry.
Install
Each server has its own README with exact commands. The pattern for Claude Code is claude mcp add, with everything after -- being the launch command. For example, the Filesystem server, published on npm as @modelcontextprotocol/server-filesystem:
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /path/to/dirFor Claude Desktop, add the same command to claude_desktop_config.json; see the Filesystem MCP Server page for a full example.
Use cases
- Learning MCP by reading small, well-scoped server implementations.
- Giving Claude Desktop file access (Filesystem) or web fetching (Fetch).
- Testing MCP clients against the Everything server, ideally with MCP Inspector.
Pros
- Maintained alongside the protocol and SDKs.
- Small, readable code that doubles as documentation.
- Free and open source.
Cons
- Limited set; most real integrations now come from service owners, like the GitHub MCP Server.
- Some overlap with Claude Code’s built-in tools (files, git, web fetch).
- Archived servers no longer receive updates, so avoid old tutorials that point to them.
Frequently asked questions
Where did the GitHub, Postgres and Puppeteer servers go?
They were archived to the modelcontextprotocol/servers-archived repository. Several have official replacements maintained by the service owner, such as GitHub's own github-mcp-server.
Is this repo a directory of all MCP servers?
No. It now holds only a small set of reference servers. To browse published servers, use the MCP Registry at registry.modelcontextprotocol.io.
Are the reference servers production-ready?
They exist to demonstrate MCP features and the official SDKs. Some, like Filesystem, Fetch and Git, are useful day to day, but review their behavior before relying on them.