MCP Databricks Server

This repository implements a Model Context Protocol (MCP) server for Databricks using the Databricks SDK. It allows LLMs or MCP clients to interact with Databricks SQL warehouses, executing queries, managing schemas/tables, and more.

Features

• Execute SQL queries
• List catalogs, schemas, and tables
• Describe table schemas
• Create schemas and tables
• Insert data into tables

Prerequisites

• Python 3.8+
• A Databricks workspace with a SQL warehouse
• A Databricks personal access token with appropriate permissions (e.g., Can Use on the warehouse)
• Ensure your SQL warehouse is running (or set to auto-start if serverless)

Setup

1. Clone the repository:

git clone https://github.com/leraz-iter/databricks-mcp.git
cd databricks-mcp

2. Create a virtual environment (recommended):

uv venv
source .venv/bin/activate

3. Install dependencies:

uv pip install -r requirements.txt

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

DATABRICKS_HOST=https://your-workspace.cloud.databricks.com
DATABRICKS_TOKEN=your-personal-access-token
DATABRICKS_WAREHOUSE_ID=your-warehouse-id

• Find DATABRICKS_HOST and DATABRICKS_WAREHOUSE_ID in your Databricks UI (SQL > Warehouses > Connection details).
• Generate a token in Databricks (User Settings > Developer > Access Tokens).

Running the Server

Run the server with:

python main.py

This starts the MCP server on stdio transport by default. For development/testing:

• Use mcp dev main.py (requires MCP CLI installed via pip install "mcp[cli]" if not already).
• Test with MCP Inspector: npx @modelcontextprotocol/inspector python main.py.

For production, consider deploying with Streamable HTTP transport (see MCP docs).

Usage with LLMs

Once running, connect via an MCP client (e.g., Claude Desktop or custom AI app). Tools like execute_sql_query can be called with natural language prompts.

Example: "Run SELECT * FROM main.default.my_table" → Invokes execute_sql_query.

Troubleshooting

• If the warehouse is stopped, start it in the Databricks UI.
• Check .env for correct credentials.
• For connection timeouts, increase polling timeout in DatabricksMCP.
• Ensure token has "Can Use" permission on the warehouse.

Security

• Never commit your .env file.
• Use least-privilege tokens.
• For production, implement MCP authentication (e.g., OAuth) as per SDK docs.

For more on MCP, see https://modelcontextprotocol.io/.