Image: GitHub
Why it mattersA coding agent that writes a web app can now log lines and stub network calls inside a real Chrome. That means less guessing when a page breaks.
A coding agent that writes a web app often cannot see what the browser does with it. Chrome DevTools added a way to check. The Chrome DevTools team released version 1.13.0 of its Model Context Protocol server. The release adds logpoints and page-scoped network mocks. A logpoint writes a message when a line of code runs. A network mock returns a pre-written response for a request. The agent can then test how the page handles that response.
The repository is ChromeDevTools/chrome-devtools-mcp. It has 55,611 stars and 4,613 forks on GitHub. The license is Apache 2.0. Trendshift's current daily list ranks the repository eighth.
What the server lets an agent do
The server runs next to the agent. It drives a real Chrome browser through Puppeteer. The agent can record performance traces and read the results. It can take screenshots and inspect network requests. It can read console messages with source-mapped stack traces. It can also navigate pages and run actions.
The Chrome DevTools team says the server officially supports Google Chrome and Chrome for Testing. The team commits to fixes for the latest Extended Stable Chrome. Node.js LTS and a current stable Chrome are required.
What 1.13.0 changed
Logpoints come from the DevTools BreakpointManager. An agent can place a logpoint on a line and read the output when the page runs. A page-scoped network mock returns a pre-written response for a single request. The release also adds upstream network mock responses. These let the agent pass a real request to the server after the mock has handled it. The test can then get both a stub and a live response.
The release fixes smaller faults too. UTF-8 byte order marks in config files now parse. Cyclic error chains stop resolving after a limit, so a bad chain cannot stop the server from responding. Binary request bodies now travel without being rewritten. The team releases a new version every two to four weeks.
Who uses it, and how
The README lists support for the main AI coding assistants. Any client that speaks the Model Context Protocol can connect. A client points at the server through one configuration entry. The server starts a browser when a tool asks for one. The project links to a tool reference document for the full list of calls.
A team building a web app needs Chrome installed on the developer's machine. The agent then watches the real browser instead of a mock. A page that renders wrong in Chrome becomes something the agent can inspect. A failing network request or a slow interaction becomes something the agent can trace and fix in the same session. The tool does not run in Firefox or Safari. The team does not say it matches the full Chrome DevTools UI.
Source
ChromeDevTools/chrome-devtools-mcp on GitHub. Version 1.13.0 release notes.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
