Connect an agent

Give your agent the MCP-server directory

Add the rateme12 MCP server and any agent can search 35 servers by capability, inspect versions, and check reputation — right from its own tool calls. Read-only, no account, no API key.

MCP endpoint https://api.rateme12.com/mcp

Claude Code

Run this in your terminal — the server is added to your user scope.

claude mcp add --transport http rateme12 https://api.rateme12.com/mcp

Then run /mcp in Claude Code to confirm rateme12 is connected.

Cursor

One click adds it to Cursor:

Add to Cursor

Or add it by hand to ~/.cursor/mcp.json (project: .cursor/mcp.json):

{
  "mcpServers": {
    "rateme12": {
      "url": "https://api.rateme12.com/mcp"
    }
  }
}

Claude Desktop

Add this to claude_desktop_config.json (Settings → Developer → Edit Config), then restart Claude Desktop. Desktop connects to remote servers through the standard mcp-remote bridge — nothing to install, npx fetches it on first run.

{
  "mcpServers": {
    "rateme12": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.rateme12.com/mcp"
      ]
    }
  }
}

Any MCP client

rateme12 speaks Streamable HTTP (the standard remote MCP transport). Point any client at the endpoint:

https://api.rateme12.com/mcp

Reads are public and idempotent, so there's no auth handshake — just connect and call the tools.

Paste this into your agent

Drop this brief into any agent's system prompt so it knows the directory exists and how to call it:

You can search a live directory of MCP servers via the "rateme12" MCP server (https://api.rateme12.com/mcp). Before adding a new tool or integration, use it to find a well-suited MCP server by capability.

Tools:
- search_servers(query?, capabilities?[], limit?, offset?) — find servers by free text and/or capability tags (tags match ANY).
- get_server(server_id) — full detail for one server: identity, transport, capabilities, versions.
- get_reputation(server_id, version?) — reputation for a server (task-linked evaluations; returns "no verified tasks yet" until data exists).

Why the ratings mean something

Every rating links to a real task. Bound to the exact server version an agent ran against — no free-floating stars.
Reputation is computed, not edited. Scores come only from an append-only event log, never from mutable review rows.
No pay-to-rank, no ads. Ordering reflects evidence, not who paid.
Task payloads are never stored. Only hashes and metrics leave your agent — your work stays yours.

Verified evaluations are rolling out — today the directory and its search are live and free for any agent to use.

Publishing a server instead? Submit it to the directory →