Claude Code gets a built-in browser that can read, click, and type on the web

Anthropic just gave Claude Code something developers have been quietly wanting for a while: a real browser. Not a simulated environment, not a limited preview — an actual tabbed browser window where the AI can read documentation, interact with issue trackers, and navigate external sites without leaving the terminal.

The feature launches as an integrated tool within Claude Code, summoned by a keyboard shortcut. It reuses the same mechanics Claude already employs to preview local applications, but Anthropic layered on safety classifiers specifically for external write actions. The browser operates from a clean profile — no cookies, no saved logins, no lingering sessions. Claude won’t purchase anything, won’t create accounts, and won’t attempt CAPTCHA bypasses unless a human explicitly signs off. That’s a sensible set of guardrails, and frankly, the kind of friction you want when an AI starts clicking around on the open web.

For enterprise teams, the controls go further. Organizations can set domain allowlists to restrict which external sites Claude can touch, or disable browser access entirely. Anthropic announced these admin levers on X, positioning them as straightforward configuration options rather than buried enterprise-only features. The company also made a clear distinction: if you want Claude working inside your own authenticated sessions — logged into your GitHub, your Jira, your internal dashboards — use the Chrome extension instead. The built-in browser isn’t designed for that use case.

What’s interesting here isn’t just the feature itself. It’s the quiet acknowledgment that developer workflows are messy and cross-platform. You’re in a terminal, then you’re in docs, then you’re copying error messages into a search bar. Letting the AI handle that context-switching directly, with safety checks that don’t slow everything to a crawl, is the kind of practical integration that actually changes how people work. The question now is how aggressively developers push against those guardrails — and whether Anthropic’s classifiers are ready for it.