MCP Servers

Connect Cursor to databases, APIs, and external tools via Model Context Protocol \u2014 extending what AI can do in your editor.

What's possible

  • Query databases directly from Chat or Composer
  • Create, update, and search GitHub issues and PRs
  • Read and write to external APIs and services
  • Access file systems, cloud storage, and documentation
  • Build custom MCP servers for your team's internal tools
  • Chain multiple MCP tools together in Agent mode workflows

Getting started

  1. 1

    Understand MCP

    Model Context Protocol (MCP) is an open standard for connecting AI tools to external systems. It lets Cursor access databases, APIs, file systems, and other tools directly. Learn more at modelcontextprotocol.io.

  2. 2

    Find an MCP server

    Browse available MCP servers for your tools — databases (PostgreSQL, SQLite), APIs (GitHub, Linear), file systems, and more. Many are open-source packages you can install via npm or pip.

  3. 3

    Configure in Cursor Settings

    Open Cursor Settings (Cmd+Shift+J) and navigate to the MCP section. Add your server configuration with the command to start it and any required environment variables.

  4. 4

    Use in Chat or Composer

    Once configured, MCP tools are available in Agent mode. Cursor can query databases, create GitHub issues, or interact with any connected service as part of its workflow.

Example use cases

Database queries

“Query the users table and show me the 10 most recently created accounts with their subscription status.”

GitHub integration

“Create a GitHub issue for the login bug we just discussed. Label it 'bug' and 'priority:high'.”

API interaction

“Check the current deployment status on our staging environment and list any failed health checks.”

Custom tooling

“Run the data migration script and show me the results. Roll back if any errors occur.”

Tips

  • Start with well-maintained MCP servers from the official registry before building custom ones.
  • Use environment variables for API keys and credentials \u2014 never hardcode secrets in MCP configurations.
  • Test MCP connections with simple queries before using them in complex Agent mode workflows.
  • MCP servers run locally on your machine. No data is sent through third-party proxies.