add-tools

To add MCP servers to a Databricks agent, update agent_server/agent.py to fetch tool definitions, add corresponding resource entries to databricks.yml with correct permissions, then deploy using databricks bundle deploy and run agent_langgraph.

Databricks agent permission errors at runtime occur when databricks.yml lacks correct resource entries or permissions for tools like Genie, vector search, or UC functions. Grant access by adding corresponding resources under resources.apps.<app>.resources.

To handle MCP tool errors and prevent agent crashes, enable the handle_tool_error option when registering tools. This returns useful error messages to the LLM instead of crashing the agent application.

For vector search and Genie spaces, add corresponding resource entries under resources.apps.<app>.resources in databricks.yml. Ensure value_from references match existing resource names to avoid misconfiguration.

Yes, you can use SQL execution and serving endpoints with Databricks MCP tools. Register them in agent_server/agent.py and grant required permissions in databricks.yml before deploying the bundle.

A common configuration pitfall when adding MCP tools to Databricks is mismatching databricks.yml value_from references to existing resource names. Ensure resource names align correctly to avoid runtime misconfiguration.