Databricks MCP Toolkit

Enterprise-grade Model Context Protocol (MCP) integration toolkit for Databricks, enabling seamless AI agent interactions with your data platform.

Overview

The Databricks MCP Toolkit provides a comprehensive implementation of the Model Context Protocol for Databricks, enabling AI agents and LLM-powered applications to programmatically interact with Databricks services. Built with enterprise requirements in mind, this toolkit offers robust cluster management, job orchestration, notebook execution, and SQL query capabilities through a standardized MCP interface.

Key Capabilities

• Cluster Operations - Full lifecycle management of Databricks compute resources
• Job Orchestration - Submit, monitor, and control Databricks job executions
• Notebook Integration - Execute notebooks with parameter passing and output retrieval
• SQL Execution - Run SQL queries against Databricks SQL warehouses
• Command Execution - Execute Python, Scala, SQL, or R code directly on clusters
• File System Access - Browse and manage DBFS storage
• Real-time Monitoring - Track execution status with automatic polling

Prerequisites

• Python: 3.10 or higher
• Databricks Access: Valid workspace URL and personal access token
• SQL Warehouse: (Optional) For SQL execution capabilities
• MCP Client: Claude Desktop, Cline, or any MCP-compatible client

Installation

Using pip (Standard Method)

# Clone the repository
git clone https://github.com/kshptl/databricks-mcp-toolkit.git
cd databricks-mcp-toolkit
# Create and activate virtual environment
python3 -m venv .venv # Use python3 on Unix/Linux/macOS, python on Windows# Activate virtual environmentsource .venv/bin/activate # Unix/Linux/macOS
.venv\Scripts\activate # Windows# Install the toolkit
pip install -e .

Using conda

# Clone the repository
git clone https://github.com/kshptl/databricks-mcp-toolkit.git
cd databricks-mcp-toolkit
# Create conda environment
conda create -n databricks-mcp python=3.10
conda activate databricks-mcp
# Install the toolkit
pip install -e .

Configuration

Create a .env file in the project root with your Databricks credentials:

# Required
DATABRICKS_HOST=https://your-workspace.azuredatabricks.net
DATABRICKS_TOKEN=dapi1234567890abcdef
# Optional (for SQL functionality)
DATABRICKS_WAREHOUSE_ID=1234567890123456

Obtaining Credentials

1. Workspace URL: Your Databricks workspace URL (e.g., https://adb-123456789.azuredatabricks.net)
2. Access Token: Generate from User Settings → Developer → Access Tokens
3. Warehouse ID: Find in SQL → SQL Warehouses → Connection Details

Using with MCP Clients

This toolkit is designed to work with MCP-compatible clients like Claude Desktop, Cline, and other AI assistants.

Claude Desktop Configuration

Add this to your Claude Desktop configuration file:

macOS/Linux: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
"mcpServers": {
"databricks": {
"command": "python",
"args": ["-m", "src.main"],
"cwd": "/absolute/path/to/databricks-mcp-toolkit",
"env": {
"DATABRICKS_HOST": "https://your-workspace.azuredatabricks.net",
"DATABRICKS_TOKEN": "dapi1234567890abcdef",
"DATABRICKS_WAREHOUSE_ID": "your_warehouse_id"
}
}
}
}

Windows Example:

{
"mcpServers": {
"databricks": {
"command": "python",
"args": ["-m", "src.main"],
"cwd": "C:\\Users\\YourName\\databricks-mcp-toolkit",
"env": {
"DATABRICKS_HOST": "https://your-workspace.azuredatabricks.net",
"DATABRICKS_TOKEN": "dapi1234567890abcdef",
"DATABRICKS_WAREHOUSE_ID": "your_warehouse_id"
}
}
}
}

Cline Configuration

In Cline settings, add the MCP server:

1. Open Cline settings
2. Navigate to MCP Servers section
3. Add new server with these settings:
   • Name: databricks
   • Command: python
   • Args: -m src.main
   • Working Directory: Path to databricks-mcp-toolkit
   • Environment Variables:
     • DATABRICKS_HOST: Your workspace URL
     • DATABRICKS_TOKEN: Your access token
     • DATABRICKS_WAREHOUSE_ID: Your warehouse ID (optional)

Verifying the Connection

After configuration, restart your MCP client and test the connection:

In Claude Desktop or Cline, try:

"List my Databricks clusters"

You should see the toolkit respond with your cluster information.

Available Commands

Once configured, you can interact with Databricks using natural language:

• "List all my Databricks clusters"
• "Create a new cluster called 'analytics' with 2 workers"
• "Run this SQL query: SELECT * FROM my_table LIMIT 10"
• "Execute this Python code on cluster xyz: df.show()"
• "Submit a notebook run for /path/to/notebook"
• "Check the status of job run 12345"

The MCP server translates these requests into the appropriate API calls.

MCP Tools Reference

The toolkit provides 23 MCP tools for Databricks operations.

Cluster Management

Tool	Description	Key Parameters
list_clusters	Enumerate all clusters	None
create_cluster	Provision new cluster	cluster_name, spark_version, node_type_id, num_workers
get_cluster	Retrieve cluster details	cluster_id
start_cluster	Start terminated cluster	cluster_id
terminate_cluster	Terminate running cluster	cluster_id

Job Management

Tool	Description	Key Parameters
list_jobs	List all jobs	None
get_job	Get job configuration	job_id
run_job	Execute existing job	job_id, notebook_params (optional)
submit_single_run	Submit one-time notebook run	run_config
get_run	Get run details	run_id
get_run_output	Retrieve run output	run_id
wait_for_run_completion	Poll until run completes	run_id, poll_interval, max_wait_seconds

Notebook Operations

Tool	Description	Key Parameters
list_notebooks	List workspace notebooks	path
export_notebook	Export notebook content	path, format (SOURCE/HTML/JUPYTER/DBC)

File System

Tool	Description	Key Parameters
list_files	Browse DBFS contents	dbfs_path

SQL Execution

Tool	Description	Key Parameters
execute_sql	Run SQL statements	statement, warehouse_id (optional), catalog, schema

Command Execution

Execute code directly on Databricks clusters with state persistence across commands.

Tool	Description	Key Parameters
create_execution_context	Initialize execution context	cluster_id, language (python/scala/sql/r)
execute_command	Run code in context	cluster_id, context_id, command, language
get_command_status	Check command status (auto-polls)	cluster_id, context_id, command_id
cancel_command	Cancel running command	cluster_id, context_id, command_id
destroy_execution_context	Clean up context	cluster_id, context_id
get_context_status	Verify context validity	cluster_id, context_id
execute_command_simple	One-shot command execution	cluster_id, command, language

Testing

# Install development dependencies
pip install -e ".[dev]"# Run all tests
pytest tests/ -v
# Run with coverage
pytest --cov=src tests/ --cov-report=term-missing

Troubleshooting

Common Issues

Connection Errors

Issue: Unable to connect to Databricks API
Solution: Verify DATABRICKS_HOST and DATABRICKS_TOKEN in .env file

SQL Execution Failures

Issue: 400 Bad Request on SQL queries
Solution: Ensure DATABRICKS_WAREHOUSE_ID is set and warehouse is running

Permission Denied

Issue: 403 Forbidden on API calls
Solution: Verify your access token has required permissions for the operation

Module Import Errors

Issue: ModuleNotFoundError
Solution: Ensure dependencies are installed: pip install -e .

Debug Mode

Enable detailed logging for troubleshooting:

LOG_LEVEL=DEBUG python -m src.main

Licensing

Dual-licensed: AGPL-3.0 or commercial. Use under AGPL-3.0 for open-source. For production use without AGPL obligations, obtain a commercial license: kushapatel@live.com.

Support

• Issues: Report bugs or request features via GitHub Issues
• Documentation: See the examples/ directory for detailed usage examples
• MCP Protocol: Learn more at Model Context Protocol

Acknowledgments

Built with the Model Context Protocol specification and Databricks Python SDK.

---

Author: Kush Patel
Repository: https://github.com/kshptl/databricks-mcp-toolkit
MCP Version: 1.0