Claude Code Gets a Built-In Browser - Read Docs, Click Through Designs, Test Sites Without Leaving Your Editor
Anthropic added a sandboxed in-app browser to Claude Code on desktop today - the day after OpenAI shut down its own standalone AI browser, Atlas.
Claude Code in-app browser launched on desktop today, giving Claude the ability to pull up any site - docs, designs, staging environments, live URLs - read it, click through it, and interact with it the same way it handles local dev servers. Anthropic's @ClaudeDevs account announced the feature on July 10 with 1.8 million views inside hours. One day after OpenAI shut down its standalone AI browser Atlas, Anthropic shipped browser capabilities directly into Claude Code.
What the Browser Does - and What It Deliberately Does Not Do
The browser runs as a pane inside Claude Code on desktop. Claude can open any URL, read the page, navigate links, click interactive elements, and test behavior - all without the developer switching windows. Anthropic's documentation describes the shortcut as Ctrl+Shift+B on Windows and Cmd+Shift+B on Mac.
Sandboxing is the key design decision. The in-app browser runs a clean profile with no saved logins, no cookies from your personal browser, and no browsing history. Developers choose whether sessions persist between runs. For reading docs, checking a design in Figma, or testing an app against a public endpoint, that isolation is the right default. Anthropic drew a clear line in its documentation: "When you want Claude to act as you in your logged-in sessions, use the Claude in Chrome extension instead, which shares your browser's login state." Two tools. Different trust levels. Different contexts.
The Opposite Architectural Bet from OpenAI
July 9 and July 10 produced mirror-image announcements from the two biggest AI coding players. OpenAI shut down Atlas - the AI-first standalone browser it launched in October 2025 - after less than nine months. OpenAI's conclusion: the browser is a feature, not a destination. So it folded agentic browsing into ChatGPT's desktop app and a new Chrome extension that competes directly with Google's Gemini Side Panel.
Anthropic went the other direction. Rather than extending an existing browser, Anthropic put a browser inside the coding tool. The bet is that developers want web context to stay in the same environment where they write and run code - not in a parallel Chrome tab or a separate app. Both decisions are defensible. OpenAI is distributing its browser reach across the platforms developers already use. Anthropic is deepening the environment developers already spend hours in. Claude Code's agentic architecture - designing loops rather than one-shot prompts - benefits from having web access inline rather than as a hand-off step.
Anthropic also has a Claude in Chrome extension for cases where Claude needs to act inside a user's authenticated browser session. That covers the logged-in-account use case OpenAI is targeting with its new Chrome extension. Between the in-app browser and the Chrome extension, Claude Cowork, and Claude Code itself, Anthropic now covers nearly every place a developer touches code or context during a working session.
Whether the sandboxed in-app approach or the Chrome-extension approach gets more daily use is the open question. Developers who run Claude Code as their primary workspace will reach for the browser pane. Developers who live in Chrome will reach for the extension. Anthropic's announcement suggesting you choose your tool based on whether you need your login state is probably the most honest framing either company has offered so far.