Databricks

You can deploy the Unity Catalog MCP server as a Databricks app. To do so, follow the instructions below:
- Move into the project directory and build the wheel:

There are two ways to deploy the server on Databricks Apps: using thedatabricks bundleCLI or using thedatabricks appsCLI. Depending on your preference, you can choose either method.

To deploy the server using thedatabricks bundleCLI, follow these steps:
- Set the env variables for theschema_full_nameandgenie_space_idsand run thebundle deploycommand:

BUNDLE_VAR_schema_full_name=catalog.schema BUNDLE_VAR_genie_space_ids=[\"space1\",\"space2\"] \ databricks bundle deploy -p your-profile-name

BUNDLE_VAR_schema_full_name=catalog.schema BUNDLE_VAR_genie_space_ids=[\"space1\",\"space2\"] \ databricks bundle run mcp-on-apps -p your-profile-name

Please note thatBUNDLE_VAR_genie_space_idsshould be exactly as shown above, with the double quotes escaped, no spaces, and the brackets included.

To connect to your app, use theStreamable HTTPtransport with the following URL:

https://your-app-url.usually.ends.with.databricksapps.com/api/mcp/

Please note the trailing/api/mcp/in the URL. This is required for the server to work correctly,including the trailing slash.

You'll also need to set theAuthorizationheader toBearer <your_token>in your client. You can get the token by running the following command:

databricks auth token -p your-profile-name

If you are a developer iterating on the server implementation, you can repeat steps #2 and #3 to push your latest modifications to the server to your Databricks app.

Please note that both variables should be provided in bothdeployandruncommands. Theschema_full_namevariable is used to determine the schema to use for the server, while thegenie_space_idsvariable is used to determine which Genie spaces to use.

To deploy the server using thedatabricks appsCLI, follow these steps:
- Move into the project directory and build the wheel:
- Configure theapp.ymlfile in the root of the project directory. You can use the following example as a starting point:

command: ["uvicorn", "databricks.labs.mcp.servers.unity_catalog.app:app"] env: - name: SCHEMA_FULL_NAME value: catalog.schema - name: GENIE_SPACE_IDS value: '["space1","space2"]'

- Deploy the app using thedatabricks appsCLI:

uv build --wheel databricks sync ./build -p <your-profile-name> /Workspace/Users/[email protected]/my-app databricks apps deploy my-app-name -p <your-profile-name> --source-code-path /Workspace/Users/[email protected]/my-app databricks apps start my-app-name -p <your-profile-name>

After the app is deployed, you can connect to it using theStreamable HTTPtransport in your MCP client, such as Claude Desktop or MCP inspector. To do this, you need to set the target URL to the URL of your app +/api/mcp/postfix. Full URL example:

https://your-app-url.usually.ends.with.databricksapps.com/api/mcp/

Please note that the URL should end with/api/mcp/(including the trailing slash), as this is required for the server to work correctly.To connect to the app, you also need to set theAuthorizationheader toBearer <your_token>, where<your_token>is the token you can get by running the following command:

databricks auth token -p your-profile-name

Please note that app service principal should be entitled with necessary permissions to access the Unity Catalog schema and Genie spaces. You can do this by assigning the appropriate permissions to the service principal in Unity Catalog and Genie.

This server is currently under construction. It is not yet usable, but contributions are welcome!

If you would like to deploy a custom MCP server on Databricks Apps, take a look at the reference implementationhere. It provides a step-by-step guide on how to create a custom MCP server and deploy it on Databricks Apps.

We welcome contributions :) - seeCONTRIBUTING.mdfor details. Please make sure to read this guide before submitting pull requests, to ensure your contribution has the best chance of being accepted.

Maritime intelligence for tracking vessels, analysing ports, and exploring ship data.