'%3e%3crect%20x='128'%20y='122'%20width='263'%20height='88'%20rx='36'%20fill='%232878ef'/%3e%3c/g%3e%3cg%20transform='rotate(7%20237%20258)'%3e%3crect%20x='128'%20y='215'%20width='216'%20height='82'%20rx='34'%20fill='%23f58220'/%3e%3c/g%3e%3cg%20transform='rotate(-5%20259%20350)'%3e%3crect%20x='128'%20y='304'%20width='264'%20height='88'%20rx='36'%20fill='%23ed438b'/%3e%3c/g%3e%3c/svg%3e){kind=small}
How agents use the built-in browser
Agents can open your site beside the conversation, read the page, click, and fill in forms to check their own work. On a Mac they can also take screenshots, type, and use the keyboard.
When an agent builds something you can see, it can test it the way you would: open the site, look at the page, and try it. This happens in the conversation’s own product browser, on the computer that runs your project.
What agents can do
On every computer, Mac, Windows, or Linux, an agent can:
- Start your site, then open it beside the conversation.
- Read the page: its text, and the buttons, links, and fields on it.
- Click buttons and links.
- Fill in text fields and choose options in lists.
- Scroll, and go to another address.
On a Mac, an agent can also:
- Take screenshots of the page to check how it looks.
- Press keys such as Enter, Tab, Escape, and the arrow keys.
- Type into rich text editors and embedded forms.
- Hover, double-click, and drag, including basic dragging on drawing canvases. Hovering works only while the browser is showing on screen.
What they can’t do
- Upload files from your computer into the page.
- Read the contents of frames embedded in a page.
- Use your computer’s clipboard, or drag and drop between apps.
- Open separate windows or download files.
- Use another conversation’s browser, or control Enjoy itself.
How it works alongside you
- It’s the same page you see. When the browser is showing, you can watch the agent click and type.
- It keeps working in the background. If you switch to another conversation or hide the browser, the agent keeps using it. It never switches your view or reopens a browser you’ve hidden.
- Pages unload when nobody needs them. When the agent finishes and you’re not looking at the page, Enjoy closes it to save memory. A page nobody has shown or used for 10 minutes also closes. The address is kept, so the page loads again the next time you or the agent need it.
- Page text is treated as information, not instructions. Agents are told not to follow instructions written on web pages.
- It shares the built-in browser’s sign-ins. If you’ve signed in to a site in the built-in browser, the agent can use that sign-in too. See Stay signed in to sites in the built-in browser.
Asking for a check
Say what to test and what a good result looks like. For example:
- “Start the site, open it, and check that the signup form shows an error for an invalid email.”
- “Open the pricing page and click every plan’s button. Tell me if any of them go to the wrong page.”
- On a Mac: “Take a screenshot of the homepage and tell me if anything overlaps.”
On Windows and Linux, agents that try to take a screenshot are told screenshots are only available on macOS, and can still read and use the page. See also Check the agent’s work.
When to use your own browser instead
If you’re already signed in to a site in Google Chrome, Claude Code can work in Chrome itself. See Let Claude Code use your Chrome browser. Your own browser is also the way to go for downloads, pages that open separate windows, and drag and drop that the built-in browser can’t do.