How to use Claude Code's built-in browser to solve real-world mysteries

Hand an AI a photo with no location data, no filename clues, nothing but pixels — and tell it to figure out exactly where it was taken. That sounds like a party trick for a detective show, not something you'd expect from a coding tool. But that's exactly the kind of task Claude Code can now pull off, because it can finally open a real browser, look at real websites, and reason through what it finds — all without leaving your workspace.

This isn't some clever prompt hack. It's a legitimate feature Anthropic shipped for Claude Code on desktop, and it changes what "AI research" actually looks like in practice. Instead of guessing based on training data from months or years ago, Claude can go check the current internet, click into pages, read documentation, and compare what it sees against your image or your code in real time.

Here's what this feature actually is, why it matters, and how to start using it today.

What is Claude Code's built-in browser?

Claude Code on desktop now has a built-in browser, and Claude can pull up docs, designs, or any other site, and read, click through, and interact with pages the same way it does with your local dev server previews. In plain terms: there's now a real browser pane living right next to your chat window, and the AI can drive it.

You can find the official release notes on the Claude Code Docs changelog and the general Claude Code desktop documentation, both of which lay out exactly what shipped and how it behaves.

The Browser pane is a tabbed browser, so you can open documentation, issue trackers, or any other site next to your running app. To open the Browser, press Cmd+Shift+B on macOS or Ctrl+Shift+B on Windows, or select it from the Views menu. That's it — no extension to install, no separate app to configure. It's built directly into the desktop client.

Why did Anthropic build this?

The gap this closes is pretty obvious once you think about it. Coding agents have always been good at reading your local files but blind to the outside world unless you copy-pasted content in manually. That sounds small, but it closes a real gap in how coding agents work — a model that can browse the web and act on what it finds is qualitatively more useful for research-heavy engineering tasks than one that can only see the files in your repository.

Practically speaking, that means Claude can now look up current library documentation instead of relying on outdated training data, check how a competitor's site actually behaves, or pull a screenshot of a bug straight from a live page — all without you switching windows.

Is it actually safe to let an AI browse the internet on its own?

This is the part Anthropic clearly spent real effort on, and it's worth understanding before you turn it loose. The browser is sandboxed and configurable: you choose whether browsing sessions persist, and safety classifiers review actions on external sites.

There's also an important distinction buried in the docs that most coverage glosses over. Anthropic's documentation explains that the Browser pane uses a clean browser profile, separate from your personal browser, with none of your saved logins or history — use it for building and testing your app and for sites that don't need your identity.

If you need Claude to interact with sites where you're already logged in — your email, a client's dashboard, whatever — that's a different tool entirely. You enable that by adding --chrome at launch or with /chrome, and its distinguishing feature is that it can directly operate logged-in web apps and localhost development servers — think of the built-in browser as a safe browser confined within the app, and Claude in Chrome as a mechanism that drives your local Chrome directly.

For organizations worried about giving an agent too much rope, there are guardrails too. Anthropic built a documented, named safety-classifier layer specific to external write actions, a separate domain allowlist gate, and two independent managed settings for organizations to dial back exactly how much of it is exposed.

How do you turn on and use the browser pane?

Getting started takes about thirty seconds if you already have Claude Code installed on desktop.

1. Update to the latest version. This shipped as part of a broader update, so make sure you're current — make sure to update to the latest version of the desktop app before looking for the feature.
2. Open the Code tab in Claude Code desktop, where your chat, files, and terminal already live.
3. Launch the Browser pane with Cmd+Shift+B on Mac or Ctrl+Shift+B on Windows — or just pick it from the Views menu.
4. Arrange your workspace. The Code tab's panes — chat, diff, browser, terminal, file, plan, tasks, subagent — can all be dragged and resized freely, so put the browser wherever makes sense for what you're doing.
5. Give Claude something to look into. Ask it to pull up a library's docs, check a live site, or research a topic, and watch it navigate on its own.

If you want to let Claude highlight something specific on a page rather than describe it in words, the Browser pane opens with Cmd+Shift+B, and elements on the current page can be selected with Cmd+Shift+S. That's genuinely useful when you're debugging a weird CSS issue and it's easier to point than explain.

What can you actually do with it beyond debugging code?

Most of the marketing around this feature focuses on developers fixing bugs and reading documentation. And that's real — Anthropic's documentation says the agent can read page content, inspect the document object model, click interface elements, fill in forms, and capture screenshots. But the feature isn't limited to code. Because Claude can search the open web and reason over what it finds, it works for open-ended research tasks too — including genuinely strange ones, like handing it a random photo and asking it to figure out where in the world it was taken based on visible clues, then cross-checking those clues against real web sources.

That's the kind of task that used to require a human doing manual reverse-image searches across multiple tabs. Now it's one prompt, with Claude doing the searching, clicking, and comparing itself.

Can it build something from what it finds?

Yes, and this is where the browser pane gets genuinely interesting instead of just convenient. Because the browser sits directly next to the coding environment, Claude doesn't have to hand research back to you as a wall of text. It can turn what it finds into an actual working interface — a webpage, a comparison tool, an interactive visualization — built from the same session where it did the research.

Practically, this means you can ask for research and a deliverable in the same conversation. Find the information, then build a small tool that presents it. No copy-pasting notes into a separate document, no context lost between "research mode" and "build mode."

What are the current limitations to know about?

The browser isn't magic, and it's worth setting expectations before you rely on it for something important.

Some sites actively block automated browsing, and if Claude reports it can't access a page or the screenshot shows a CAPTCHA or paywall, that's a real limitation of the approach, not a bug — for those cases, you'll need to manually copy the relevant content into your prompt.

Also worth knowing: Anthropic's Computer Use API is a separate, more powerful capability that lets Claude control an entire desktop environment, while the in-app browser in Claude Code is narrower and focused on browser-based tasks within your development workflow — for most developers, the in-app browser is what you actually need day-to-day. Don't confuse the two when you're deciding what tool fits your task.

And if authenticated access matters for what you're doing — logging into an account, checking something behind a paywall — remember the clean-profile limitation. For most research and documentation tasks, authentication isn't an issue, but know the limitation. You'll want Claude in Chrome for anything that requires your actual logged-in session.

Where should you look for more?

Start with Anthropic's own Claude Code documentation for the full technical rundown on panes, permissions, and settings. The Week 28 changelog is the original source for exactly what shipped and when. And if you want a deeper look at practical use cases beyond debugging — things like competitive research and visual UI comparisons — MindStudio's breakdown walks through several patterns worth stealing.

The bigger shift here isn't really about browsers or keyboard shortcuts. It's that AI tools are quietly moving from "answer questions based on what I already know" to "go find out, then show your work." Once an agent can browse, click, verify, and then build something from what it learns — all in one sitting — the line between researching a problem and solving it starts to disappear. Try handing your own AI something weird and open-ended this week. You might be surprised what comes back.