AI agent tools

AI agent tools give your agents practical capabilities beyond text generation, like searching documents, querying databases, calling REST APIs, or running custom code.

Use pre-configured managed MCP servers for immediate access to Databricks data, use external MCP servers to connect to third-party APIs, or build custom tools for specialized business logic.

Tools and examples

Common tool patterns and implementation examples:

• • Work with structured data
  • Read structured data in Unity Catalog tables.
• • Retrieve unstructured data
  • Connect agents to vector search indexes to query unstructured data.
• • Code interpreter tools
  • Let agents run Python code dynamically using the built-in system.ai.python_exec tool.
• • AI tools using Unity Catalog functions
  • Create tools using Unity Catalog functions. (Databricks recommends using MCP tools instead for most new use cases.)

MCP servers

Use MCP servers to give agents access to governed Databricks data and third-party APIs:

• • Connect agents to external services
  • Use managed OAuth, the UC connections proxy, or external Rest APIs to connect agents to third-party APIs.
• • Use external MCP servers in agents
  • Call external MCP servers from agent code through Databricks-managed proxies. To install an external MCP server, see Install an external MCP server.
• • Use custom MCP servers in agents
  • Connect agent code to a custom MCP server hosted as a Databricks app. To host a custom MCP server, see Host a custom MCP server.