Genie MCP Server

This project provides an async server for interacting with Databricks Genie via a MCP StreamableHTTP Protocol. It enables users to query Genie and receive answers or data, leveraging Databricks authentication and robust error handling.

Features

• Async API server using Starlette and Uvicorn
• Integration with Databricks Genie for conversational queries
• Automatic OAuth token management for Databricks

Project Structure

• mcp_server.py: Main server entry point, exposes the tool via MCP StreamableHTTP Protocol.
• genie_room.py: Handles Genie API interactions and conversation logic.
• token_minter.py: Manages Databricks OAuth token minting and refreshing.
• requirements.txt: Python dependencies.
• app.yaml: Example deployment configuration.

Deploying to Databricks Apps

You can deploy the Genie MCP Server as a Databricks app by following these steps:

1. Clone the Repository to Your Workspace

   In your Databricks workspace, navigate to the directory where you want to deploy the app (e.g., /Workspace/Users/your.email@databricks.com/genie_mcp_server). Then, clone the repository:

   git clone https://github.com/your-org/genie_mcp_server.git
   

2. Configure the Genie Space ID and Other Environment Variables

   Open the app.yaml file in the root of the cloned repository. Update the SPACE_ID value to match your Genie space.

   Example app.yaml:

   command:
   - "python"
   - "mcp_server.py"
   
   env:
   - name: "SPACE_ID"
     value: "your_space_id"
   

3. Create and Deploy the App in Databricks

   • Go to the Databricks Apps interface.
   • Create a new app and specify the path to the directory where you cloned the repository.
   • Complete the app creation and deployment process. databricks apps deploy genie-mcp-server --source-code-path /Workspace/Users/your.email@databricks.com/genie_mcp_server

4. Access the API

   Once deployed, your Genie MCP Server app will be running and accessible via the configured port and endpoint (e.g., https://app-url.aws.databricksapps.com/api/mcp/). You can now send MCP requests to your deployed app from within your Databricks workspace or from external clients, depending on your network configuration.