Databricks integrated MCP server governance and tool controls into Unity Catalog and the AI Gateway.

Databricks expanded support for the Model Context Protocol (MCP) across its governance and application stack during August 2026. Unity Catalog added governance coverage for MCP servers, tools, and agent workflows, introducing centralized discovery, fine-grained access policies, and audit logging to manage tool access [3]. Alongside this foundation, technical tutorials for Agent Bricks and Databricks Apps demonstrated how teams can integrate MCP servers into custom agents while tracking runtime steps through MLflow execution traces [5, 6, 7].

Administrative controls also matured through additions to the Databricks AI Gateway. The interface allowed platform administrators to review MCP server configurations and filter which specific tools could be exposed to automated agents [2]. The gateway also enabled automated usage tracking to monitor token consumption and spending across connected agent tools [2]. Outside Databricks tooling, Anthropic highlighted MCP as a standard primitive for production agents running autonomous troubleshooting workflows [4].

Practical deployment hurdles surfaced in external client environments. Practitioners working with Anthropic Claude workspaces reported that setting up multiple Databricks Genie MCP connectors caused them to collapse into a single shared connector [1]. This behavior created tool-routing issues for teams attempting to connect separate Genie data sources within the same workspace [1].