Databricks
Connect Databricks' managed MCP servers so agents can ask questions of a Genie space and call Unity Catalog functions, under each user's own Databricks permissions.
Databricks runs its own managed MCP servers, and Airia's catalog includes them. Instead of configuring a separate query tool for every table, you point a server at something you've already built in Databricks, such as a Genie space or a Unity Catalog schema of functions. Agents then use whatever that space or schema covers, and Databricks applies Unity Catalog permissions to every call.
The Databricks servers are catalog servers marked Available, so they show up automatically for organizations in Default Mode. In Custom Mode, an admin approves them on the Server Management page before they can be added to a Gateway or Deployment.
Which Server to Use
| Server | What agents can do | You need |
|---|---|---|
| Databricks (Genie) | Ask natural-language questions about the business data in one Genie space and get answers grounded in that space's data model. Read-only. | A Genie space and its Space ID |
| Databricks (Unity Catalog) | Call the Unity Catalog functions in one catalog and schema as tools. | A catalog and schema that contain the functions |
| Databricks | Call Databricks' built-in AI functions in system.ai, such as classification, extraction, and translation. | A workspace |
For questions about your data across many tables, Genie is usually the place to start. The Genie space holds the tables, joins, and instructions, so you curate the data model once in Databricks and every agent connected to the space uses it.
To cover several Genie spaces or schemas, add one connection per space or schema to the same Gateway or Deployment.
Before You Connect
Have a Databricks account admin create an OAuth app connection for Airia. Airia's connection isn't registered automatically, so this one-time step comes first.
Sign in to the Databricks account console (accounts.cloud.databricks.com for AWS and GCP workspaces, accounts.azuredatabricks.net for Azure) and go to Settings → App connections → Add connection.
Enter a name, such as Airia MCP Gateway, and add this redirect URL:
https://auth.airia.ai/OAuth/callbackLeave Generate a client secret checked. Under Access scopes, select All APIs.
Airia asks for a narrower scope when each person signs in (genie offline_access for Genie, unity-catalog offline_access for the other two). Databricks grants that scope only to a connection that has All APIs, and offline_access keeps people signed in without re-authenticating.
Click Add, then copy the Client ID and Client Secret from the Connection created dialog. Databricks shows the secret only once.
Connection Details
Add the server to a Gateway or Deployment, enter the Client ID and Client Secret, and fill in the fields below. Each person then clicks Connect and signs in with their own Databricks user, so they see only the data their Unity Catalog grants allow.
Databricks (Genie)
| Field | What to enter |
|---|---|
| Workspace | Your workspace hostname up to, but not including, .databricks.com. For dbc-a1b2c3d4-e5f6.cloud.databricks.com, enter dbc-a1b2c3d4-e5f6.cloud. |
| Space ID | Open the Genie space in your browser and copy the part of the URL after /genie/rooms/. It's also on the space's Configure → About tab. |
Databricks (Unity Catalog)
| Field | What to enter |
|---|---|
| Workspace | Your workspace hostname up to, but not including, .databricks.com, as for Genie. |
| Catalog | The catalog that holds the functions. Browse catalogs under Catalog in the workspace sidebar. |
| Schema | The schema within that catalog. Each function in it becomes a tool. |
For workspaces on Azure Databricks (adb-….azuredatabricks.net), use Databricks (Genie) - Azure and Databricks (Unity Catalog) - Azure from the catalog.
Databricks
| Field | What to enter |
|---|---|
| Workspace | The full workspace hostname without https://, such as dbc-a1b2c3d4-e5f6.cloud.databricks.com or adb-1234567890123456.7.azuredatabricks.net. |
This server also accepts a personal access token instead of OAuth. Generate one in the workspace under Settings → Developer → Access tokens → Generate new token and paste it in as the API key. The token carries the permissions of the user who created it.
Troubleshooting
Learn More
- Databricks managed MCP servers
- Connect MCPs to AI assistants and coding agents
- Enable or disable partner OAuth applications
Related Resources
Tenant vs. Personal Level App Credentials
Decide whether a Databricks credential is shared or personal
Server Management
Approve which MCP servers your organization can use
Gateway/Deployment Creation
Add Databricks to the Gateway or Deployment you're building
Snowflake
Connect Snowflake's managed MCP server the same way