Chrome DevTools MCP get_css_styles: agents read real CSS

The most common way a coding agent fails at front-end work is simple: it edits CSS it cannot see render. Google's latest Chrome DevTools MCP release closes part of that gap. A new get_css_styles tool lets a coding agent read the computed and matched styles of an element on a live page, the same data you see in the Styles and Computed panes. Google lists it first in its November "New in DevTools" post, which covers Chrome 155 and 156.

What actually happened

From Google's post and the chrome-devtools-mcp repository:

• get_css_styles: returns computed and matched styles, including inherited rules, pseudo-elements, @keyframes and at-rules such as @position-try, @property and @function. It pages results with pageSize and pageIdx, so a large stylesheet does not flood the agent's context.
• Config file and path controls: server flags can now live in a dedicated configuration file, and you can set custom file system roots. The CLI uses --allow-unrestricted-paths by default.
• Bigger traces: a chunked parser and a 1.2 GB default trace buffer let agents read large performance recordings.
• Fewer stuck runs: JavaScript dialogs that pop up during input actions are handled automatically, and console history survives same-document navigations.
• Setup: the DevTools main menu now has a "Get Chrome DevTools for agents" item that opens the setup steps.

For humans, the same release adds a visual position-area editor for CSS anchor positioning and diagnostics for Web Install manifests.

Why it matters for your business

Before this, an agent fixing "the button overlaps the footer" read the source CSS and guessed which rule won. On a real storefront with a theme, an app's injected styles and three media queries, that guess is often wrong. The agent then makes a second edit, and a third. Each loop costs tokens and review time. With computed and matched styles, the agent can see which rule actually applies, then change that one.

One flag needs your attention. Google's notes say the CLI now uses --allow-unrestricted-paths by default, and Google does not explain the scope in the post. Read the flag's documentation in the repo before you connect an agent, and assume the widest reading: file access outside your project folder. On a developer laptop with client repos, SSH keys and .env files, that is too wide. Set explicit file system roots in the new config file and commit it to the repo, so every developer and every agent runs with the same limits.

Sources: Chrome for Developers: New in DevTools, ChromeDevTools/chrome-devtools-mcp on GitHub.