Claude in Chrome
Anthropic's Chrome extension that lets Claude read, click, type and fill in forms in your own browser, with the tabs and logins you already have, started from a side panel, Cowork or Claude Code.
My take
I've used it to extract data and to control web apps for me. At Carewell I've used it to create and update the product documentation of our web app and to run tests on it. I even had Claude talk to our customer service chatbot: AI testing AI. Like the Office add-ins, it isn't really connected to my usual work setup, and most agent tools (Cowork, Claude Code, Codex, Gemini and others) have their own browser. So tech-savvy users will probably go that way. But it has its charm, because it's so simple.
At a Glance
Pros
- + Nothing to set up: install the extension, and Claude works in the browser you already use
- + Works on pages you're signed in to, including internal tools and old portals that have no connector
- + Can use your own web app the way a person would, which makes it useful for documentation and testing
- + Side panel sessions are saved to your Claude account and can be continued in the Claude apps
- + Saved shortcuts and scheduled tasks for things you repeat
Cons
- − Hidden instructions on a web page can steer it, and Anthropic itself calls it risky
- − Everything visible in the tab becomes part of the conversation
- − Chrome only, and not on phones
- − If you already work in Claude Code, Cowork or Codex, their built-in browsers may suit you better
Best for: People who want an agent to work on websites without setting anything up: pulling data out of web apps, filling in forms, clicking through tools they're already signed in to
Full review
Claude in Chrome is Anthropic’s extension for Google Chrome. It lets Claude work in your own browser: it reads the page you’re on, clicks, types, fills in forms and moves between tabs, using the logins you already have. You start it from a side panel in Chrome, or let Cowork or Claude Code use it.
What I use it for
I’ve used it to get data out of web pages and web apps, and to operate web apps for me, which means Claude clicks and types where I otherwise would.
At Carewell I’ve used it on our own web app, to create and update its product documentation and to run tests. And I had Claude hold conversations with our customer service chatbot, so one AI was testing another. If you run a chatbot, it’s a simple way to have it questioned by something other than your own team.
How it works
Open the side panel on any tab and describe the task. By default it runs in “Automatically approve” mode: Claude checks each action itself and stops to ask you when something needs your approval, and purchases and other steps you can’t take back wait for you. If you’d rather see every action, switch to “Manually approve”.
A few things make repeated work easier. You can drag tabs into Claude’s tab group so it works across them. Shortcuts are saved prompts you call with /, and you can schedule them; scheduled tasks need Chrome to stay open.
According to Anthropic, the side panel now runs a Cowork session: it’s saved to your account, you can continue it on the web, in the desktop app or on your phone, and the skills and connectors on your account work there too. Your files stay on your computer: tasks that need them only work while the Claude desktop app is open and connected.
The other route: the browser built into your agent
Most agent tools now bring their own browser. Cowork and Claude Code can drive your Chrome through this extension, but in the Claude desktop app they also have a browser pane of their own, with its own logins, kept apart from your Chrome. And an agent can run a headless browser: a real browser, often the same one Chrome is built on, running without a window on any screen. Nobody looks at it. The agent drives it with code: it loads pages, clicks, reads them and takes screenshots. This site uses one: when Claude Code changes a page, it can open it in a headless browser and check it at phone and desktop sizes.
If you already work in Claude Code, Cowork or Codex, you’ll probably reach for their browsers. They stay out of your own browser and its logins. Claude in Chrome’s charm is its simplicity: install an extension, open a side panel, say what you want. No terminal, no setup.
Before you use it
Anthropic says it plainly: Claude in Chrome is “enhanced with our safety classifiers but is still risky.” The main risk is hidden instructions on a web page that try to steer Claude into doing something you didn’t ask for. Claude also works from screenshots of the tab, so anything visible becomes part of the conversation, and side panel sessions are saved to your history.
Anthropic’s advice: use a separate Chrome profile that isn’t signed in to your bank, health or government accounts, and start with sites you trust. Claude asks before it goes onto financial sites and blocks adult and pirated content, and Anthropic advises against using it for financial accounts, legal documents and health information.
At work, your company decides whether you can use it. Admins on Team and Enterprise plans can switch it off for everyone or limit the sites it may open. If your company hasn’t approved it, don’t use it on work sites with a private Claude account: whatever is on the page goes into your conversation. I used it on our own web app. If you want to do the same, check with whoever runs the app first, and use a test account rather than one that sees real customer data.
Who should use Claude in Chrome
People who want an agent to do things on websites and don’t want to install or configure anything. It’s also a good first look at what an agent does when it uses a computer the way you do. If you already work in Claude Code, Cowork or Codex, you’ll probably reach for their browsers instead.
Added: 2026-10-18 · Last updated: 2026-10-18