← back
Building Agent Interfaces: Lessons from Chrome DevTools (MCP) for Agents — Michael Hablich, Google
Takeaway
Design tools around agent reasoning limits and measure successful outcomes rather than simply minimizing tool count or tokens.
Summary
- Chrome DevTools for agents provides browser debugging and performance-tracing tools through MCP-compatible harnesses.
- Raw traces containing tens of thousands of JSON lines overwhelmed agents, motivating semantic Markdown summaries of actionable metrics.
- Interface evaluation separates task completion from efficiency and measures tokens per successful outcome within comparable task classes.
- Optional tool categories and a three-tool slim mode reduce context overhead, with tradeoffs in extra agent turns and available capabilities.
chrome-devtoolsmcpagent-interface
Original description
Chrome DevTools MCP shipped with one tool: debug_webpage. Agents failed silently because they couldn't compose behaviors. The team decomposed it into 25 focused tools and assumed the problem was solved. It wasn't — now agents had 25 tools and no reliable way to pick the right one. Michael Hablich's talk is an honest account of building the same thing wrong three times and what the fixes actually looked like.
The concrete lessons: semantic summaries instead of raw 50,000 line JSON trace files, error messages rewritten so agents can self heal without a human in the loop ("Cannot navigate back, no previous page in history" instead of "Unable to navigate back in currently selected page"), a metric called tokens per successful outcome to measure interface fuel efficiency, and a deliberate decision to keep the autoconnect friction rather than remove it once they thought through prompt injection and the lethal trifecta.
Speaker info:
https://x.com/MHablich
/ michael-hablich