Cursor’s AI agent can now open a browser, take screenshots, and use what it sees to fix UI bugs — no copy-pasting error messages or describing layout problems in words. Version 1.7 shipped browser controls alongside two other features that change how the agent works: Plan Mode for reviewing before executing, and Hooks for running custom scripts automatically.
What happened
Three additions landed in Cursor 1.7:
Browser controls. The agent can launch a browser, navigate to a local dev server, take screenshots, and inspect the result. When it spots a broken layout or a rendering bug, it can iterate — edit code, reload, screenshot again — without the developer manually checking each change. This closes a feedback loop that previously required constant alt-tabbing between Cursor and a browser.
Plan Mode. Before the agent starts writing code, you can now ask it to produce a plan first. You review the plan, adjust it, then approve execution. This is Cursor’s answer to a common complaint: agents that sprint ahead and make sweeping changes before you understand what they’re doing.
Hooks. These are user-defined scripts that run automatically at specific points in the agent’s workflow — after a file is saved, before a command runs, or when the agent finishes a task. Think auto-formatting, linting, test runs, or deployment triggers baked directly into the agent loop.
TIP
Hooks pair well with Plan Mode. Set up a Hook to run your test suite after every agent edit, then use Plan Mode to review the approach before the agent starts. You get both a preview of intent and automatic verification of results.
Why it matters
Browser controls solve a specific, persistent friction point in AI-assisted frontend work. Until now, every AI coding tool — Cursor included — was effectively blind to what the UI actually looked like. The agent could read your React components and CSS, but it couldn’t see that a button was overlapping a sidebar or that a modal wasn’t centered. Developers had to describe visual problems in text, which is slow and lossy.
With screenshots in the loop, the agent gets direct visual feedback. This is particularly useful for iterative UI polish: adjusting spacing, fixing responsive breakpoints, catching rendering regressions. The agent can verify its own work visually instead of relying on the developer to confirm each change.
Plan Mode addresses a trust problem. Agents that modify 15 files in one shot are powerful but nerve-wracking, especially in unfamiliar codebases. Seeing the plan first lets developers catch wrong assumptions before they propagate across the project.
WARNING
Browser controls currently work with local dev servers. If your app requires authentication flows, complex environment setup, or backend services to render properly, the agent may not be able to reach a meaningful screenshot without additional configuration.
Who should care
Frontend developers using Cursor daily will get the most from browser controls — anyone building UIs who’s tired of describing visual bugs in chat. Full-stack developers working on admin dashboards, landing pages, or component libraries will see immediate time savings on the design-tweak-verify cycle.
Plan Mode matters most for developers working in larger codebases or on teams where unexpected file changes create merge conflicts. Hooks are for developers who already have linting, formatting, or test scripts and want them wired into the agent’s workflow without manual triggers.
Practical takeaway
Update to Cursor 1.7, enable browser controls in settings, and try it on a frontend task you’d normally eyeball manually. Start your next agent session with Plan Mode on — review what the agent intends before it touches code. If you have a formatter or linter you run after every edit, set it up as a Hook so the agent’s output stays clean automatically.