Databricks MCP Server

--- description: "Set up and deploy a Databricks MCP server from this template" --- # 🚀 Databricks MCP Server Setup I'll help you set up your own Databricks MCP (Model Context Protocol) server so Claude can interact with your Databricks workspace! ## 📋 What We'll Do ⏺ **Step 1: Environment Setup** - Configure Databricks authentication ⏺ **Step 2: Deploy MCP Server** - Deploy to Databricks Apps ⏺ **Step 3: Add to Claude** - Configure Claude to use your MCP server ⏺ **Step 4: Test Connection** - Verify everything works ⏺ **Step 5: Customize** - Add your own prompts and tools (optional) --- ## Step 0: Check Configuration and Existing MCP Setup **First, let me check your MCP server configuration:** [I'll read config.yaml to get your server name] ```yaml # config.yaml servername: [your-server-name] # Default is 'databricks-mcp' ``` **Now let me check if your MCP server is already set up:** ```bash # Check if MCP server is already added to Claude SERVER_NAME="[server-name-from-config]" echo "list your mcp servers" | claude | grep -q "$SERVER_NAME" && echo "✅ MCP server '$SERVER_NAME' already configured!" || echo "❌ MCP server '$SERVER_NAME' not found" ``` [I'll run this check with the actual server name] **If the MCP server is already configured AND .env.local exists:** ✅ Your MCP server '$SERVER_NAME' is already set up! What would you like to do? 1. **Skip to customization** - Jump to Step 5 to add custom prompts and tools 2. **Test the connection** - Go to Step 4 to verify everything works 3. **Redeploy with changes** - Continue with Step 2 to update deployment 4. **Start fresh** - Remove and reinstall: `claude mcp remove $SERVER_NAME` 5. **Continue from the beginning** - Go through all steps **Please let me know which option you'd prefer!** **If .env.local doesn't exist:** - 🆕 This looks like a fresh clone - let's start from Step 1! --- ## Step 1: Environment Setup **Let me check if your environment is already configured:** [I'll check for .env.local file existence] ```bash # Check if .env.local exists if [ -f ".env.local" ]; then echo "✅ Great! .env.local found - environment is already configured" # Test authentication source .env.local && export DATABRICKS_HOST && export DATABRICKS_TOKEN && databricks current-user me else echo "📋 Starting fresh - let's create your .env.local with setup" fi ``` **If .env.local doesn't exist, I'll run the interactive setup script:** ```bash # Open a new terminal for interactive setup CURRENT_DIR=$(pwd) if [ -d "/Applications/iTerm.app" ]; then osascript <<EOF tell application "iTerm" create window with default profile tell current session of current window write text "cd $CURRENT_DIR && ./setup.sh --auto-close" end tell activate end tell EOF else osascript <<EOF tell application "Terminal" do script "cd $CURRENT_DIR && ./setup.sh --auto-close" activate end tell EOF fi ``` [I'll wait for you to complete the setup and say "done" when you're ready to continue] **Please say "done" when the setup script completes!** --- ## Step 2: Deploy MCP Server **After you've said "done", I'll deploy your MCP server:** **a) Check if app exists:** [I'll run `./app_status.sh` to check current status] **b) Create and deploy app:** ```bash # Deploy (creates app if needed) nohup ./deploy.sh --create --verbose > /tmp/mcp-deploy.log 2>&1 & ``` **c) Monitor deployment:** [I'll tail the log file and monitor progress] ```bash # Monitor deployment log tail -f /tmp/mcp-deploy.log ``` **d) Wait for app to be fully ready:** [I'll keep checking app status until it's RUNNING] ```bash # Keep checking until app is ready while true; do STATUS=$(./app_status.sh | grep "App Status:" | awk '{print $NF}') if [ "$STATUS" = "RUNNING" ]; then echo "✅ App is now RUNNING!" break else echo "⏳ App status: $STATUS - waiting..." sleep 10 fi done ``` **e) Get your app URL:** [Once the app is RUNNING, I'll show you the deployed app] ```bash # Get and display app URL export DATABRICKS_APP_URL=$(./app_status.sh | grep "App URL" | awk '{print $NF}') echo " ✅ Your MCP server is deployed! 🌐 App URL: $DATABRICKS_APP_URL 🔗 You can visit your app at: $DATABRICKS_APP_URL " ``` --- ## Step 3: Add MCP Server to Claude **Only after the app is fully deployed and RUNNING, we'll add it to Claude:** **Now let's add your MCP server to Claude!** [I'll read the configuration from .env.local and app_status.sh] ```bash # Get configuration source .env.local export DATABRICKS_APP_URL=$(./app_status.sh | grep "App URL" | awk '{print $NF}') # Add MCP server to Claude claude mcp add $SERVER_NAME --scope user -- \ uvx --from git+ssh://git@github.com/databricks-solutions/custom-mcp-databricks-app.git \ dba-mcp-proxy \ --databricks-host $DATABRICKS_HOST \ --databricks-app-url $DATABRICKS_APP_URL ``` [I'll execute this command with your actual values] --- ## Step 4: Test Connection **Let's verify your MCP server is working:** ```bash # Test with echo trick echo "What MCP prompts are available from databricks-mcp?" | claude ``` [I'll run this and show you the results] **Expected output:** - Should list available prompts (check_system, list_files, ping_google) - Should show available tools (execute_parameterized_sql, etc.) **Note:** You'll need to restart this Claude session to see the MCP server in the `/mcp` command. The echo test confirms it's working for new sessions. --- ## 🎉 Success! **Your Databricks MCP server is successfully deployed!** 🔄 **Important:** Please restart Claude to see your MCP server in the `/mcp` list. **Would you like to work on adding custom tools or prompts?** I can help you: - 🛠️ Add custom tools for specific Databricks operations - 📝 Create custom prompts for your workflows - 🚀 Both tools and prompts - ✅ No thanks, I'm all set! **What would you like to do?** --- ## Step 5: Customize (Optional) ### Add Custom Prompts Create markdown files in the `prompts/` directory: ```markdown # prompts/my_custom_prompt.md # Description of what this prompt does Content that will be returned to Claude ``` ### Add Custom Tools Add functions in `server/app.py`: ```python @mcp_server.tool def my_custom_tool(param: str) -> dict: """Tool description for Claude.""" # Your implementation return {"result": "data"} ``` [Details on how to add custom prompts and tools will be shown based on your choice above]