databricks-agent-bricks

Use the manage_ka tool with action create_or_update, providing a name and a volume_path pointing to documents in a Unity Catalog Volume. The endpoint provisions in 2-5 minutes, and example questions from companion JSON files are added automatically.

A Supervisor Agent (formerly Multi-Agent Supervisor) routes user queries to specialized agents such as Knowledge Assistants, Genie Spaces, model serving endpoints, UC functions, or external MCP servers. Each agent needs a name, a routing description, and exactly one source identifier.

Yes, a Supervisor Agent can mix agent types in one configuration. Provide genie_space_id for SQL-based data agents and ka_tile_id for document Q&A agents, with clear descriptions so the supervisor routes queries correctly.

Use manage_ka with action find_by_name and the exact KA name. It returns the tile_id, sanitized name, endpoint_name, and endpoint_status, which you can then use when adding the KA to a Supervisor Agent.

Provisioning typically takes 2-5 minutes, but can take up to 10. If it stays in PROVISIONING longer, check workspace capacity and quotas, verify the volume path is accessible, and confirm file permissions on the Unity Catalog Volume.

Create a Unity Catalog HTTP Connection with is_mcp_connection set to true pointing at your JSON-RPC 2.0 endpoint, grant the agent service principal USE CONNECTION, then reference it in the agent configuration via the connection_name field.