A Model Context Protocol (MCP) server that connects to Databricks API, allowing LLMs to run SQL queries, list jobs, and get job status.

• Run SQL queries on Databricks SQL warehouses
• List all Databricks jobs
• Get status of specific Databricks jobs
• Get detailed information about Databricks jobs

• Python 3.7+
• Databricks workspace with:
  • Personal access token
  • SQL warehouse endpoint
  • Permissions to run queries and access jobs

1. Clone this repository
2. Create and activate a virtual environment (recommended):

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Create a .env file in the root directory with the following variables:

   DATABRICKS_HOST=your-databricks-instance.cloud.databricks.com
   DATABRICKS_TOKEN=your-personal-access-token
   DATABRICKS_HTTP_PATH=/sql/1.0/warehouses/your-warehouse-id
   

5. Test your connection (optional but recommended):

   python test_connection.py
   

1. Host: Your Databricks instance URL (e.g., your-instance.cloud.databricks.com)
2. Token: Create a personal access token in Databricks:
   • Go to User Settings (click your username in the top right)
   • Select "Developer" tab
   • Click "Manage" under "Access tokens"
   • Generate a new token, and save it immediately
3. HTTP Path: For your SQL warehouse:
   • Go to SQL Warehouses in Databricks
   • Select your warehouse
   • Find the connection details and copy the HTTP Path

Start the MCP server:

python main.py

You can test the MCP server using the inspector by running

npx @modelcontextprotocol/inspector python3 main.py

If you are integrating this server with an MCP client (like Cursor), you might configure it using Docker. The client will typically manage running the Docker container based on a configuration file (e.g., mcp.json).

A pre-built image is available on Docker Hub and can be pulled using:

docker pull jordineil/databricks-mcp-server

The configuration passes environment variables directly to the Docker container. Here's an example structure, replacing placeholders with your actual credentials and using the public image name:

{
  "mcpServers": {
    "databricks-docker": {
      "command": "docker",
      "args": [
        "run", 
        "--rm", 
        "-i", 
        "-e",
        "DATABRICKS_HOST=",
        "-e",
        "DATABRICKS_TOKEN=",
        "-e",
        "DATABRICKS_HTTP_PATH=",
        "jordineil/databricks-mcp-server" 
      ]
    }
    // ... other servers ...
  }
}

• Replace with your Databricks host (e.g., dbc-xyz.cloud.databricks.com).
• Replace with your personal access token.
• Replace with the HTTP path for your SQL warehouse.

This method avoids storing secrets directly in a .env file within the project, as the MCP client injects them at runtime.

The following MCP tools are available:

1. run_sql_query(sql: str) - Execute SQL queries on your Databricks SQL warehouse
2. list_jobs() - List all Databricks jobs in your workspace
3. get_job_status(job_id: int) - Get the status of a specific Databricks job by ID
4. get_job_details(job_id: int) - Get detailed information about a specific Databricks job

When used with LLMs that support the MCP protocol, this server enables natural language interaction with your Databricks environment:

• "Show me all tables in the database"
• "Run a query to count records in the customer table"
• "List all my Databricks jobs"
• "Check the status of job #123"
• "Show me details about job #456"

• Ensure your Databricks host is correct and doesn't include https:// prefix
• Check that your SQL warehouse is running and accessible
• Verify your personal access token has the necessary permissions
• Run the included test script: python test_connection.py

• Your Databricks personal access token provides direct access to your workspace
• Secure your .env file and never commit it to version control
• Consider using Databricks token with appropriate permission scopes only
• Run this server in a secure environment