Databricks MCP Server Template

You can host Databricks MCP prompts and tools as a dedicated server that Claude can talk to. This setup creates a bridge between Claude and your Databricks workspace, enabling secure authentication, prompt loading, and tool execution through a managed MCP endpoint.

How to use

You interact with your MCP server from Claude or Claude CLI. After you deploy and connect your server, Claude can list available prompts and tools, and you can invoke prompts to get information or run Databricks actions through your MCP-backed API.

Prompts are loaded from the prompts directory and exposed to Claude as named prompts. Tools are Python functions decorated to be available as MCP tools. When you call a prompt or tool from Claude, your server authenticates securely, translates requests, executes the underlying Databricks actions, and returns structured results.

How to install

Prerequisites: you should have Python and an environment suitable for running a FastAPI/MCP server, plus any required development tooling for your Databricks integration.

Step 1. Prepare your MCP server from the template setup

Step 2. Install and configure dependencies

Step 3. Deploy to Databricks Apps and verify the deployment

Configuration and usage notes

You configure the MCP server to connect to your Databricks workspace and Databricks Apps. The server will load prompts from the prompts/ directory and expose tools defined in server/app.py. For local development and testing, you can run a development proxy that authenticates with OAuth through Databricks Apps.

Key environment variables used during setup and operation include host URLs and identifiers used by Claude to connect to your MCP server.

Examples of deployment and usage steps

# Deploy to Databricks Apps
./deploy.sh

# Check status and get your app URL
./app_status.sh

# Deploy to Databricks Apps
./deploy.sh

# Check status and get your app URL
./app_status.sh

Available tools

list_clusters

Lists Databricks clusters by status and returns a JSON-serializable summary of cluster IDs, names, and states.

create_job

Creates a Databricks job with a given name, notebook path, and cluster, returning the job ID and a link to run the job.