About mcp.tc
A directory of MCP servers. Every server gets a page at a link you can guess, with setup that connects you straight to it.
What mcp.tc is
mcp.tc is a directory of Model Context Protocol servers: the services and local programs that give AI apps such as Claude, ChatGPT, Cursor or VS Code extra tools. It lists 536 of them right now.
Every listing gets a link built from the server’s name, like mcp.tc/i/notion or mcp.tc/i/deepwiki. The page behind it says what the server does, which tools it has, whether it needs a sign-in, and how to add it to each client. The link is easy to paste into a chat or a README.
How the quick link works
The quick link is the server’s page. It says what the server does, lists its tools and gives you one-click setup for Claude, Cursor, VS Code and other clients. Every button, command and config snippet on it uses the server’s own URL or install command, so your client connects straight to the server. mcp.tc isn’t in between: your tool calls, sign-in tokens and API keys never reach us.
That’s on purpose. You shouldn’t have to trust a middleman with what your AI app sends to a server, and every extra hop is one more thing that can break. Sign-in couldn’t pass through one anyway: an OAuth token only works at the server’s own URL.
When an MCP client opens the link
The quick link isn’t a server address. A client that tries to connect to it gets an error message that names the server’s own URL or install command.
Renamed and removed listings
When a listing is renamed, its old link sends browsers to the new one. When a listing is removed, the link answers that it’s gone.
How a server gets listed
Anyone can suggest a server by pasting its URL into the submit form. There’s no account. We also add well-known servers from a curated list and from the official MCP Registry.
- A robot connects to the server the way a client would: it runs the MCP handshake, asks for the tool list and checks how sign-in works. It also reads the server card, the registry entry and the GitHub, npm or PyPI page when they exist.
- An AI model (Claude, run by Anthropic) reads that public metadata and drafts the listing: name, summary, category, what you can do and a few example prompts. It runs with no tools, so instructions hidden in a manifest can at most change the draft, and the draft still has to pass the fixed checks below.
- Fixed checks then decide. A listing goes live on its own only when the server answered, the reviewer approved it, nothing was flagged and the name matches who runs it. Everything else waits for a person.
Listed servers are checked again every day, so health stays current. When a server’s tool list changes, the new list is published after a person reviews it. The terms list the exact criteria.
What mcp.tc doesn’t do
We don’t run, audit or vouch for the servers listed here. A listing means the server answered and looked like what it claims to be, nothing more. Connecting an MCP server gives an AI app new abilities, and a badly written or hostile server can misuse them. Read a server’s tool list before you connect it, and prefer servers from vendors you already trust.
There are no accounts, comments, ratings or popularity counts here. Google Analytics runs only if you accept it, and nothing else on the site tracks you.
For machines
Every listing is also available as Markdown at /i/{name}.md. There’s an llms.txt, an Atom feed of new servers, a sitemap and a small search index at /api/index.json. The API docs cover them all, along with an API in the MCP Registry’s format, and mcp.tc’s own MCP server lets your assistant search the directory.
Who runs it
mcp.tc is run by Marco Costantino, based in Italy. It isn’t affiliated with Anthropic, the MCP project or any of the vendors listed. Write to privacy@mcp.tc with corrections, questions or removal requests, or use the report page.