modelcontextprotocol/servers
The reference servers maintained with the protocol: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time, with older servers moved to an archive.
A Model Context Protocol server gives an agent access to a system it cannot reach on its own: a browser, a database, a billing API. A skill teaches the method; a server supplies the tools. The best setups pair the two.
Claude Code reads servers from .mcp.json in a project or from a plugin's mcpServers entry; Cursor, Codex, Gemini CLI, Goose and Grok Build have their own configuration files for the same servers. The official registry publishes a server.json for each listed server with a verified namespace.
A server holds credentials and acts on outside systems, so check what tools it exposes and what scopes its keys carry before you add it. The directory lists servers from their publisher's repository or official documentation and does not score or scan them.
Add the server to your agent's configuration. In Claude Code, claude mcp add registers a command or URL; a project can also commit a .mcp.json file so every collaborator gets the same servers.
claude mcp add <name> -- npx -y <package>
claude mcp add --transport http <name> <url>
Documentation: Model Context Protocol · Official MCP registry · MCP in Claude Code · Skills vs MCP servers vs plugins
Where the protocol's own servers and the community lists live.
The reference servers maintained with the protocol: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time, with older servers moved to an archive.
The official registry: publishers describe a server in server.json, verify their namespace through GitHub or DNS, and publish with the mcp-publisher CLI.
The largest community list of MCP servers, grouped by category, with a companion directory at glama.ai.
Official servers that show up in most setups.
Drives a live Chrome through DevTools: navigate, inspect the DOM, read console and network activity, record performance traces.
Browser automation through Playwright's accessibility tree, so an agent can fill forms, click and read pages without screenshots.
Pulls current, version-specific library documentation and examples into the agent's context.
Connect to a publisher-operated service using its documented authentication and usage model.
Retrieve public profiles, posts, videos and conversations for research through Scrape Atlas’s hosted MCP. Requires a Scrape Atlas account and a compatible MCP client; use the publisher’s documented OAuth or customer-key setup. Successful retrievals, including partial results, consume account credits; paid access uses one-time credit top-ups. Follow source-platform terms and data-use permissions.
Reviewed by the directory's editors, each pairing a server with the skills that know how to use it.
60 listed skills with full pages connect to an MCP server or name MCP tools in their files. Browse them with their scores.
Entries are added by hand from their publisher’s repositories or official documentation and described from what the publisher documents. They carry no Suede Skill Check score, no security result and no verification badge: the score applies to SKILL.md skills only. To suggest a repository or a correction, open a listing request.
A program that speaks the Model Context Protocol and exposes tools an agent can call, resources it can read and prompts it can use. It usually holds the credentials for an outside service, such as a database or a billing API.
Use a skill when the agent has the access it needs and is missing the method. Use a server when the job needs data or actions the agent cannot reach with files and a shell. Many featured projects in this directory pair a server with the skills that know how to use it.
Run claude mcp add with the server's command or URL, or commit a .mcp.json file in the project. Plugins can also declare servers, so one install brings the skills and the tools together.
No. Servers are listed from their publisher's repository or official documentation with a short description. They carry no quality score, no security result and no verification badge from the directory.