Connect an MCP server to Cursor

Cursor reads MCP servers from a JSON file — one per project, or one for everything. You can also click through Settings → MCP → Add new MCP server; it writes to the same file.

Where the config lives

Remote (hosted) server

Vendor-run servers need only a URL — Cursor handles the sign-in flow:

{
  "mcpServers": {
    "linear": {
      "url": "https://mcp.linear.app/mcp"
    }
  }
}

Ready to paste right now: fal.ai, Atlassian, Datadog, Buffer, Google Drive — each page has the exact URL.

Local (stdio) server

Local servers are launched by Cursor as a subprocess:

{
  "mcpServers": {
    "example": {
      "command": "npx",
      "args": ["-y", "@vendor/mcp-server"],
      "env": { "VENDOR_API_KEY": "sk-…" }
    }
  }
}

Python servers use uvx as the command. Package names, arguments and required environment variables come from each server's README, which is linked from its verdict page here.

Using it

Enable the server in Settings → MCP — a green dot means it connected and its tools were listed. Then just describe what you want in the agent panel; Cursor picks tools on its own, or you can name the server directly. Tool calls ask for approval unless you turn on auto-run, which is worth leaving off for anything that writes.

When the dot stays red

Questions