Databricks

You learn how to host MCP prompts and tools for Claude on Databricks Apps, creating a secure, authenticated bridge between Claude and your Databricks workspace. This server lets you expose prompts and Python-based tools that interact with Databricks APIs, so your AI assistant can query systems and run actions inside your workspace with clear, controllable access.

How to use

You interact with your MCP server through Claude once it is deployed. Start by deploying your server to Databricks Apps, then connect Claude to your deployed endpoint. You will see a list of available prompts and tools that you defined, and you can call them from Claude to retrieve information or perform actions in your workspace. The prompts provide descriptive content, while tools execute defined operations such as listing clusters or creating jobs. Authentication is handled securely by the proxy, so you don’t need to manage OAuth flow manually during normal use.

How to install

Prerequisites you need before you begin: Python installed on your machine, and the Databricks CLI configured with your workspace credentials. You should also have Git and the GitHub CLI available for cloning templates and scaffolding your server.

# Preinstall checks
python --version
pip --version
git --version
gh --version

# Clone your MCP server project from a template or repository
# Example placeholder for cloning steps
# git clone <your-mcp-repo-url>
# cd <your-repo-directory>

# Install Python dependencies
pip install -r requirements.txt

# Optional: set up a virtual environment
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Configuration and deployment

After you have your project files ready, you deploy the MCP server to Databricks Apps and obtain the app URL. You will use this URL to connect Claude to your MCP server.

# Deploy to Databricks Apps (example command flow)
./deploy.sh

# Check deployment status and retrieve the MCP URL
./app_status.sh

Environment and startup notes

Key environment variables you use during development and deployment include the host and app URLs that Claude needs to reach your MCP server, as well as credentials for Databricks OAuth handled by the proxy in production.

Available tools

list_clusters

List Databricks clusters by status and return basic details such as cluster_id, cluster_name, and state.

create_job

Create a Databricks job with a given name, notebook_path, and cluster_id, then return the job_id and a run URL.

dynamic_status

Get dynamic system status information from the workspace and return a formatted status string.