Chrome DevTools MCP

Chrome DevTools MCP connects a coding agent to a real Chrome instance, through the Model Context Protocol or the command line, so the agent can inspect a live page, read console and network activity, and run Lighthouse audits. Google's framing is that it lets an agent validate its own front-end work against a browser rather than against its assumptions.

That is a specific fix for a specific failure. An agent writing front-end code works from a description of the DOM rather than the DOM: it produces code that would work if the page were as it imagined, and reports success on that basis. Handing it a browser puts the feedback loop back on reality — the element really is not where it thought, the fetch really did fail CORS, the layout really does shift. The measurement is not the agent's own opinion of its output, which is the failure mode this class of tool exists to remove. The caveat is scope: a real browser is a real attack surface, and an agent driving one with your session and network access can reach things you did not intend.