✦ MCP SERVER DIRECTORY

Find the right MCP server

Browse, search and compare 5,950+ Model Context Protocol servers from the Official MCP Registry and curated awesome-lists — with a ready-to-run install command for each one.

Frequently asked questions

About & data sources →

What is an MCP server?

An MCP server is a program that exposes tools, resources and prompts to an AI application over the Model Context Protocol, an open JSON-RPC 2.0 based protocol. A server can run locally as a subprocess (stdio transport) or be reached over HTTPS as a remote endpoint (Streamable HTTP transport). The current protocol revision is 2026-07-28.

Where does mcpnav's data come from?

The data comes from the Official MCP Registry (registry.modelcontextprotocol.io), whose terms dedicate published server metadata to the public domain under CC0 1.0, plus two MIT-licensed curated lists: punkpeye/awesome-mcp-servers and wong2/awesome-mcp-servers. Repository metadata such as stars and archive status is read from GitHub. mcpnav does not copy README files and does not scrape other directory sites.

Why doesn't every server in the Registry have a page here?

The Registry is self-published and permissive by design; its maintainers say consumers should assume minimal-to-no moderation. mcpnav generates a page only for records that are not archived on GitHub and that either appear in one of the two curated lists, have at least 10 GitHub stars, or carry an official flag from those lists. The remaining records stay in the dataset but are not rendered.

How do I get my server listed?

Either publish it to the MCP Registry with the mcp-publisher CLI (the Registry's quickstart walks through server.json, namespace verification and `mcp-publisher publish`), or open a pull request adding it to punkpeye/awesome-mcp-servers, which accepts contributions through its CONTRIBUTING guide. Note that wong2/awesome-mcp-servers states in its README that it does not accept pull requests. mcpnav re-reads all three sources on each refresh; there is no separate submission form.

How is the Connect configuration on a server page generated?

It is generated from the server's Registry metadata. A `packages` entry with a stdio transport becomes an `npx`, `uvx` or `docker run` command and a matching JSON `mcpServers` snippet; a `remotes` entry becomes a `claude mcp add --transport http` line and JSON with the endpoint URL. Every environment variable or header the Registry marks as required or secret is shown by name with a `<YOUR_VALUE>` placeholder; mcpnav never fills in real values.

What is the difference between a remote and a local server?

A local server is a process your client launches on your machine and talks to over stdin/stdout; it runs with your privileges and your data stays local unless the server itself makes network calls. A remote server is an HTTPS endpoint you send requests to; whatever you send leaves your machine, and you authenticate with a header or OAuth. The Streamable HTTP spec requires remote servers to validate the Origin header and says they should authenticate all connections.

How often is the data updated?

Data is refreshed automatically once a day: an incremental pull from the Official MCP Registry (using its updated_since filter) plus a re-check of GitHub metadata for every listed server, followed by a rebuild and deploy. Remote endpoint reachability is re-probed weekly. Each page's data is taken from the most recent fetch, and the sitemap's lastmod reflects that fetch date. This follows the Registry's own guidance for aggregators: fetch on a regular but infrequent schedule.

Is mcpnav affiliated with Anthropic or the Model Context Protocol project?

No. mcpnav is an independent directory. It consumes public data from the Official MCP Registry and two community-maintained lists, and it is not operated by, endorsed by, or affiliated with Anthropic, the MCP project, or the Registry maintainers. Servers appear here because their metadata is public, not because anyone reviewed them.