datasource-mcp

MySQL datasource MCP server for AI clients (Cursor, Claude Desktop, etc.).

Database Support: Currently supports MySQL only. Support for PostgreSQL, SQLite, and other databases is planned for future releases.

Install

pip install datasource-mcp

If the package is not yet available on your PyPI mirror, use the official index:

pip install datasource-mcp -i https://pypi.org/simple/

Or run without installing (requires uv):

uvx datasource-mcp

Note: uvx datasource-mcp downloads the package and immediately starts the MCP stdio server. It will block the terminal while waiting for a client connection. Configure it in your MCP client instead of running it manually.

Related MCP server: mcp-db-server

Cursor / MCP Client Config

Option 1: pip install (local environment)

Run pip install datasource-mcp first, then add this to mcp.json:

{
  "mcpServers": {
    "datasource-mcp": {
      "command": "datasource-mcp",
      "env": {
        "DB_HOST": "127.0.0.1",
        "DB_PORT": "3306",
        "DB_USER": "root",
        "DB_PASSWORD": "your_password",
        "DB_NAME": "your_database"
      }
    }
  }
}

Option 2: uvx without installing (recommended)

No pip install required — uvx fetches the package from PyPI and runs it automatically. Requires uv to be installed.

{
  "mcpServers": {
    "datasource-mcp": {
      "command": "uvx",
      "args": ["datasource-mcp"],
      "env": {
        "DB_HOST": "127.0.0.1",
        "DB_PORT": "3306",
        "DB_USER": "root",
        "DB_PASSWORD": "your_password",
        "DB_NAME": "your_database"
      }
    }
  }
}

After updating the config, click Refresh in the MCP panel to apply changes.

Environment Variables

Tools

db_exe

Execute SQL and return a JSON string:

{"success": true, "row_count": 1, "truncated": false, "rows": [...]}

Error response:

{"success": false, "error": "error message"}

Local Development

uv sync
uv run datasource-mcp
# or
uv run python -m datasource_mcp

Requires Python >= 3.11.