UX/UI Radar Číst česky
Tool-stack and workflow ideas

Idea – Give your agent Chrome DevTools instead of screenshots

When an agent reads real computed styles through DevTools, it stops guessing from a screenshot. Try it on one component that differs from the design and see the difference in 15 minutes.

The change we suggest: when debugging UI in code, connect Chrome DevTools for agents to your agent and let it work with the page's real styles.

Why now

The same pattern keeps repeating: tools give agents direct access to real state. The Figma agent reads prototype connections when creating code layers (release notes of 3 Nov 2026), Framer's Harness 2 entry says the same capabilities are available to external agents through MCP, and DevTools now offers get_css_styles with computed and matched styles, @keyframes or @property rules. So the idea rests on a trend, not a single announcement.

What you get

  • The agent sees which rule won the cascade instead of guessing from an image.
  • A gap between design and implementation (spacing, color, font) is found from numbers.
  • The output can go straight into a fix or a comment for developers.

Limitation: the Chrome post gives no performance or pricing details, and how the connection behaves with your agent you must verify yourself.

First step, 15 minutes

  1. In DevTools, open the main menu and choose Get Chrome DevTools for agents.
  2. Follow the setup to connect it to your agent.
  3. Pick a button that differs from Figma and ask: "Explain where this button's padding and color come from and propose a minimal change."
  4. Compare the answer with the Styles tab.

Sources