Theo predicted this months ago: coding agents belong in the browser, where your UI already lives. React Grab for Agents makes that real by sending prompts to the agents you already use, like Claude Code, Cursor or Copilot, straight from the page.
I’ve been using it on inbox.run and my side project Inbound, and it has changed how quickly I ship UI fixes.
Summary
- Add one script to a Next.js or other React project. Hold ⌘C and click an element to copy its component stack with exact file paths and line numbers, not a screenshot.
- With that context, the agent opens the right file instead of searching for it. On a shadcn dashboard benchmark, React Grab made Claude Code roughly 66% faster on UI tasks.
- In agent mode, you select an element, press Enter and type a change. A local Claude Code instance edits the file while status updates stream back to the page.
- You can run several agents on the page at once for parallel UI tasks.
- It connects to your existing agent subscriptions through open-source providers, at no extra cost.
How to apply it
- Add React Grab to your dev script, and use ⌘C selection whenever you describe UI to an agent.
- Keep prompts small and specific. A precise target also keeps the agent’s context usage low.
- Use agent mode for quick visual tweaks during reviews or calls with clients.
- Run separate agents on separate elements to batch UI fixes.
Verdict
The fastest way I’ve found to make precise UI edits with an agent. It’s React-only and still early, but I use it every day.
What’s good
- Exact component context instead of vague descriptions
- Edits without leaving the browser
- Works with the agents and subscriptions you already have
What’s not
- React-only, since it reads the React component tree
- Early: I hit an agent error during the demo
Get the next test the day it’s out.
