If you’ve been using Claude to handle web tasks—research, form-filling, booking, data entry—you’ve probably noticed the awkward middle step: launching a separate browser, copying links, managing logins, and hoping the AI doesn’t lose context halfway through. Anthropic just removed that friction. Starting this week, Claude Cowork on desktop comes with its own built-in browser, so the AI can open sites, read pages, click buttons, and type into forms without ever touching your personal tabs or passwords.

The update lands inside the Claude desktop app for macOS, Windows, and Linux. When a task requires a website, a Chromium-based browser opens in Cowork’s side panel instead of your main browser. Claude navigates, interacts, and completes the job while you keep working in your own window.

Crucially, this browser runs with a clean profile. It doesn’t inherit your cookies, saved passwords, or extensions. If a site needs you to be logged in, you bring that session over explicitly—site by site—rather than handing Claude blanket access to everything.

Anthropic says the feature is rolling out over the coming week to Pro, Max, Team, and Enterprise plans. There’s nothing to install and no extension to enable; it’s baked into the desktop app.

For power users, this closes a long-standing gap. Before, you either relied on the “Claude in Chrome” extension—which mirrors your active tab and shares your logged-in state—or you pasted links and screenshots into chat and hoped Claude could infer what to do. The extension model works, but it blurs boundaries: your AI assistant sees whatever you’re seeing, including sensitive dashboards and personal accounts.

The built-in browser flips that. It gives Claude a dedicated workspace for public-facing tasks—research, documentation, form submissions, multi-step workflows—while keeping your primary browser and its credentials separate. Think of it as a sandboxed “agent browser” that lives next to your real one.

How does it work?

Imagine you ask Claude to compile a list of project management tools, visit each vendor’s pricing page, and fill out a demo request form with your company details. With the new setup, Claude opens those sites in the side panel, reads the content, clicks through pricing tiers, and completes forms—without pulling in your Gmail, Slack, or banking cookies.

If a site requires authentication, you can import your login for that domain only. Anthropic frames this as a deliberate, per-site choice rather than a global handover. The company also notes a familiar caveat: even with isolation, prompt-injection attacks on web pages remain a risk, so users should still treat untrusted sites carefully.

Anthropic has been explicit about guardrails. The Cowork browser doesn’t see your other tabs, history, or saved passwords unless you explicitly import a session for a specific site. It also doesn’t stay logged in everywhere by default; you decide which domains get your credentials.

This approach aligns with how Cowork already handles local files: you grant access to specific folders, actions are logged, and you can stop operations at any time. The same principle applies here—narrow permissions, transparent logs, and user control.

This isn’t Anthropic’s first major desktop upgrade this year. In April, the company rebuilt the Claude Code experience around parallel sessions, adding a “Mission Control” sidebar, integrated terminal, in-app editor, and drag-and-drop panes. The message was clear: the desktop app is becoming a full workflow environment, not just a chat window.

The built-in browser extends that vision to non-coding tasks. Between persistent threads, project workspaces, and now a sandboxed browser, Cowork is positioning itself as a place where multi-step, cross-app work happens—without constant context switching.

The feature is rolling out to paid plans: Pro, Max, Team, and Enterprise. It’s available on macOS, Windows, and Linux (the latter in beta). If you’re on a free plan, you won’t see it yet; Anthropic typically gates advanced automation features behind paid tiers.

For individuals, the biggest win is friction reduction. You can hand off a web task and watch it unfold in the side panel without juggling tabs or worrying about credential leakage. For teams, the isolation model makes it easier to delegate research and data-entry workflows without exposing everyone’s logged-in accounts.

It also sharpens the line between “agent browsing” and “human browsing.” Your primary browser stays yours—bookmarks, extensions, personal sessions intact—while Claude gets a clean, purpose-built environment for executing tasks.

Anthropic’s move reflects a broader trend: AI assistants are evolving from chatbots into action-takers. The bottleneck has rarely been intelligence; it’s been access. By giving Claude a first-class browser inside the desktop app, Anthropic removes one of the last major barriers to end-to-end automation on the web.

Whether you’re compiling competitive research, onboarding to new tools, or automating repetitive web forms, the built-in browser turns “here’s a link, can you…?” into “done, here’s what I found.” And it does so without asking you to surrender your entire digital life to the AI.

If you’re on a paid plan and use the desktop app, keep an eye on updates over the next week. The browser should appear automatically. When it does, try handing Claude a multi-site task and watch it work in the side panel—no extension, no setup, and no shared logins unless you choose.