Official MCP Server Catalog

✓ Official★ 88.372

The official reference repo for Model Context Protocol servers — the starting point for almost every MCP setup.

The Model Context Protocol (MCP) is an open standard that gives LLMs secure, controlled access to external tools and data sources — without every vendor inventing its own proprietary plugin format. This repository is the official collection of reference implementations, maintained by Anthropic together with the community.

It currently contains seven reference servers: Everything (a test server with all MCP features), Fetch (retrieve web content), Filesystem, Git, Memory (persistent context), Sequential Thinking, and Time. Older servers that used to live in the main repo — for example GitHub, Postgres, or Brave Search — have moved to the separate servers-archived repository or are now maintained directly by the respective vendor.

TypeScript servers can be started directly via npx without installation, Python servers via uvx or the classic pip install. Each server is then registered in the mcpServers configuration of Claude Desktop or Claude Code (command + arguments), after which its tools become available to the agent. The repo also points to the MCP Registry (registry.modelcontextprotocol.io), where community servers can be browsed — the protocol itself is now being developed further within a broader governance structure involving several major vendors, no longer by Anthropic alone.

🔧 SETUP

Pick the right server in the repo (e.g. `server-memory`, `server-filesystem`, `server-git`)
Start a TypeScript server without installation: `npx -y @modelcontextprotocol/server-memory`
Start a Python server via `uvx mcp-server-git` (alternatively `pip install mcp-server-git && python -m mcp_server_git`)
Add the configuration in Claude Desktop/Code, e.g. `{"mcpServers": {"memory": {"command": "npx", "args": ["-y", "@modelcontextprotocol/server-memory"]}}}`
Look for further community servers via the MCP Registry (registry.modelcontextprotocol.io)

✅ WHEN TO USE IT?

Getting started with MCP

Anyone setting up an MCP server for the first time will find the official, tested reference implementations here as a template.

Building your own server

The reference servers (e.g. Filesystem, Git) serve as code templates for writing your own MCP server for a custom data source.

⛔ WHEN NOT TO?

Looking for specific popular servers (GitHub, Postgres, Brave Search)

These were moved from the main repo to the archive (servers-archived) or are now maintained directly by the respective vendor.

Better: Search directly in the vendor's repo (e.g. github/github-mcp-server) or in the MCP Registry.