Claude in Chrome is Anthropic’s experimental extension for letting Claude see and act inside your actual browser. Rather than spinning up a clean automated browser, it works in the Chrome profile you already use, with the sessions you are already logged into. That is what makes it powerful and what makes it worth being careful with.
What it actually does
The extension gives Claude a view of the current page and the ability to act on it: read content, click, fill fields, and move between pages. Because it runs in your own profile, it can reach applications behind a login without you scripting an authentication flow, which is the step that usually kills browser automation for everyday tasks.
Practical patterns:
- ‘Read this dashboard and summarise what changed since last week.’
- ‘Fill this form using the details from the document we just discussed.’
- ‘Go through these ten tabs and tell me which ones are worth keeping.‘
Why use it
Most browser automation tools are built for testing: deterministic, headless, starting from nothing. That is the wrong shape for “help me with the thing I am looking at”. Working inside a real session means the agent sees what you see, which suits research, admin and the sort of small repetitive web tasks nobody has ever bothered to script.
Gotchas
The authenticated session is the whole risk. An agent operating as you can do anything you can do, and a web page it reads can contain text written specifically to hijack an agent reading it. Prompt injection stops being theoretical the moment the browser is logged in. Enable it for a defined task, watch what it does, and turn it off afterwards. It has also been an experimental product with shifting behaviour, so verify current capability against Anthropic’s own docs rather than any third-party write-up, this one included.
