zozo123 opened a new issue, #74193: URL: https://github.com/apache/airflow/issues/74193
### Description Add a Databricks-provider adapter that resolves a Unity MCP service and uses the existing generic MCP toolset machinery. Keep Databricks authentication and service addressing in the Databricks provider, with an optional dependency on common.ai. ### Use case/motivation A DAG author should be able to give an in-worker agent governed tools using a Databricks connection and a service name, without copying bearer tokens or gateway URLs into DAG code. One MCP integration can expose services such as UC functions, Genie and AI Search where those services are configured and authorized. ### Proposed scope and acceptance - Reuse existing Databricks connection authentication; document the caller identity and supported auth modes. - Validate service names and construct URLs within the configured workspace; never forward credentials to an arbitrary caller-supplied host. - Refresh credentials across long-lived calls/reconnections; retain generic transport behavior in common.ai. - Translate access denial, missing service, throttling and transport errors distinctly. Do not blindly retry mutating tool calls after ambiguous delivery. - Keep common.ai optional: importing unrelated Databricks operators must continue to work without AI dependencies. - Test authentication refresh, URL construction, denied access, optional-dependency behavior and tool discovery. - Provide an opt-in system test: AgentOperator → configured Unity MCP service → list tools → harmless tool call. Record actual live validation separately from mock tests. Start with invocation of an existing service. Service administration/CRUD and SDK modernization are separate follow-ups. ### Related issues Independent of #74191; does not block or expand that PR. References: https://docs.databricks.com/aws/en/unity-gateway/concepts and https://airflow.apache.org/docs/apache-airflow-providers-common-ai/stable/index.html -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
