Databricks MCP Workshop

Interactive workshop for learning how to build agents on Databricks with hands-on examples covering managed servers, external integrations, custom server development and local IDE setup.

🎓 Workshop Participant Setup

For workshop participants - Get your own isolated environment with one command:

git clone https://github.com/databricks-solutions/mcp-workshop
cd mcp-workshop
./setup.sh

This interactive setup creates your personal workshop environment:

• 🏗️ Your own Unity Catalog with sample data
• 🚀 Custom MCP server deployed as a Databricks App
• 💻 Local development environment
• 🔐 Secure authentication setup

Each participant gets isolated resources (e.g., mcp_workshop_john_doe) - no conflicts, no shared data issues!

👉 See detailed setup guide →

🏢 Enterprise/Admin-Managed Setup

For workshops where participants don’t have catalog creation permissions:

1. Admin: Run setup/admin_bulk_setup.py notebook → Admin Guide
2. Participants: Follow the quick start → Participant Guide

Each participant gets their own isolated schema within a shared catalog - no jobs, no catalog creation needed!

---

Access the Workshop

After deployment, access the workshop locally:

1. Navigate to the frontend: cd frontend
2. Start the dev server: npm run dev
3. Visit http://localhost:3000 to start the workshop

Your MCP server will be deployed as a Databricks App (check the Apps page)

📚 Workshop Content

Section 1: Managed MCP Servers (45 min)

Learn to use Databricks-managed MCP servers with:

• Unity Catalog functions and permissions
• Vector Search for RAG applications
• Genie Spaces for natural language SQL
• Automated vs Interactive modes

Section 2: External MCP Servers (40 min)

Integrate third-party services:

• Unity Catalog connections
• GitHub, Atlassian, Slack integrations
• Custom API connections
• Proxy authentication and security

Section 3: Custom MCP Servers (60 min)

Build and deploy your own:

• Simple @mcp.tool() decorator pattern (based on databrickslabs/mcp)
• Databricks SDK integration (clusters, warehouses, SQL)
• Markdown-based prompts for reusable commands
• Databricks Apps deployment with automated scripts
• End-to-end testing and debugging tools

Section 4: Local IDE Integration (30 min)

Connect your development environment:

• Cursor IDE configuration
• Claude Desktop setup
• VS Code with MCP extensions
• OAuth and PAT authentication

🔐 Authentication Model

This workshop uses Databricks Apps built-in authentication exclusively:

• User Authorization: Your access token is automatically forwarded via x-forwarded-access-token header
• App Authorization: Service principal credentials via DATABRICKS_CLIENT_ID and DATABRICKS_CLIENT_SECRET environment variables