Unity Gateway

Unity Gateway works like a checkpoint that every external tool call from an AI agent must pass through. It's a bit like installing CCTV cameras at every door of a building — later, just by reviewing the footage, you can tell who went through which door and when, or whether someone got turned away because a door was locked. When an agent calls an outside service such as Jira or Google Drive using the common protocol (Model Context Protocol, MCP), the gateway relays that call and automatically records which tool was invoked, what values were passed, whether an error occurred, how long it took, and how many tokens were used.

The key point is that these logs are always being accumulated. There's no need to add separate monitoring code just to hunt for problems. Because the gateway already sits at the point where every call passes through, if you later suspect that costs suddenly spiked or that an agent is behaving oddly, you can simply look back through the logs that have already been collected.

The article explains that "Databricks built Genie One on top of Unity Gateway, which automatically logs every MCP call... and used it to uncover a hidden bug." A common misunderstanding: Unity Gateway itself is not a tool that finds or fixes bugs. The gateway only records call logs — a separate query tool was used to ask natural-language questions against those logs and extract the answer.