Reviewed by Jonathan West · Updated Aug 29, 2026

Claude in Chrome Can Click, Type, and Fill Forms for You

The extension reached general availability on August 26, 2026, and it works on the sites you are already signed into.

Reviewed by Jonathan West · Updated Aug 29, 2026

Claude in Chrome is a browser extension that can see the page you're on and interact with it using your existing logins. In the systems we build for clients, the browser is often the last mile of automation, the work lives behind a login, with no application programming interface (API) to access it.

Anthropic made the extension available to all paid Claude plans on August 26, 2026. It can read and type text, click links, navigate between pages, and fill out forms.

That makes it useful for a specific kind of task: work your team still has to do manually in a browser because the system has no usable API. You can now hand that work to Claude and watch it happen.


What Claude in Chrome Does

Claude in Chrome views the page you are on and acts on it, using your existing logins (Anthropic). It reads and types text, clicks links, navigates between pages, and fills out forms.

The login detail is the part that changes what is possible. A connector needs a vendor to publish an API and needs someone to wire it up. The extension needs neither, because it works the way you do, inside a session you already opened.

It can also take actions on its own rather than asking you to approve each one (Anthropic). That turns a click-by-click assistant into something that finishes a task while you do other work.

  • Reads and types text on the page, so it can pull a figure out of a dashboard or fill a field you would otherwise type by hand.
  • Clicks links and navigates between pages, so a task that spans four screens does not need four separate prompts.
  • Fills out forms using your existing logins, which is what makes it work on internal tools that were never built for automation.
  • Acts autonomously when you let it, instead of stopping for approval on every single action.
The extension works on systems with no API. That is the whole reason to reach for it over a connector.

Not sure whether your browser work needs Claude in Chrome or a proper API integration? At Layer3Labs, we map which of your systems publish an API and which are stuck behind a login.

Book a Consultation

Who Can Use It and What It Costs

Claude in Chrome is generally available on every paid Claude plan (Anthropic). There is no separate purchase and no waitlist.

Anthropic does not publish a price for the extension itself, because it is bundled with the plan you already pay for. Check the current plan tiers on Anthropic's pricing page before you budget for seats, since plan contents change without notice.

You install it from the Chrome Web Store like any other extension. The requirement worth planning around is the browser itself: it runs in Google Chrome, and not everywhere else.


How the Permission Model Works

Every action Claude in Chrome takes passes a safety classifier before it runs, which checks that the action is safe and matches what you asked for (Anthropic). The check sits between the model deciding to click and the click happening.

The extension also screens web content for prompt injections using what Anthropic calls probes. A prompt injection is text hidden on a page that tells the model to do something you never asked for. A browser agent reading live web pages is the target it was invented for.

By default Claude approves the actions it judges safe on its own. You can turn that off in settings so every action waits for you (Anthropic). Teams handling money, customer records, or anything that sends email should turn it off on day one and leave it off until they have watched the thing work.

  • Safety classifier: validates each action before it is performed against what you actually requested.
  • Injection probes: screen page content for instructions planted by a site rather than written by you.
  • Auto-approval: on by default for actions Claude judges safe, and switchable in settings.
  • Manual mode: every action waits for a human click, which is the right setting for a first rollout.

What It Cannot Do Yet

Claude in Chrome does not run on other Chromium browsers or on mobile (Anthropic). If your team standardized on Edge, Brave, or Arc, the extension is not an option for them today.

That single limit decides more rollouts than the capability list does. A five-person team on mixed browsers has to move everyone to Google Chrome first, and that is a change-management project rather than an install.

Mobile is the other gap. Work that happens on a phone stays manual for now, so a field team that lives in a mobile browser gets nothing from this release.

Check which browser your team runs before you plan anything around the extension. Mixed-browser teams have a migration to run first.

When the Extension Beats an API Integration

Reach for Claude in Chrome when the system you need has no API, or when the API exists but does not expose the field you care about. That is the case it was built for and the case where nothing else works.

Across the workflows we have automated for small and mid-sized business teams, the jobs that survive longest as manual work sit behind a login. A vendor portal. An insurer's claims screen. A state filing site, or an internal tool nobody has budget to extend. Every one of those is a browser tab someone opens by hand.

Where a real API does exist, build against the API instead. It is faster, it does not break when a vendor moves a button, and it runs without a human logged in. A browser agent is what you reach for on the systems an API never covered.

  • Use the extension: vendor portals, legacy internal tools, government filing sites, anything with a login and no API.
  • Use an API integration: your customer relationship management (CRM) system, your accounting platform, your help desk, and anything else with published endpoints.
  • Watch for breakage either way: a page redesign breaks a browser agent the same week it ships.

Who Should Not Install It

Skip Claude in Chrome if your team is not on Google Chrome, because it does not run on other Chromium browsers or on mobile (Anthropic). There is no workaround, and waiting costs you nothing.

Skip it as well if the browser work touches money movement, patient records, or anything a regulator audits, and no one can watch it run. The permission controls are real, but a reviewed action is only reviewed if someone reviews it.

In both cases the better next step is the same. Write down the five browser tasks that eat the most hours. Then check which of those systems publish an API, because those are cheaper to automate properly than to automate by clicking.


What Would Change This Answer

Support for other Chromium browsers would remove the largest blocker, since it is the browser requirement rather than the capabilities that stops most teams. Anthropic's wording says the extension does not run on them yet, so the limit reads as timing.

Mobile support would open a second set of jobs entirely, particularly for teams whose work happens away from a desk.

A published enterprise control surface would change the answer for regulated buyers. Central policy over which sites the extension may touch, and an audit log of what it did, is what a compliance team asks for first.

Until then the next step is small. Install Claude in Chrome on one machine, leave auto-approval off, and point it at a single browser task you already do by hand.

Frequently Asked Questions

  • Yes. Claude in Chrome is generally available on every paid Claude plan, and you install it from the Chrome Web Store like any other extension (Anthropic). It does not run on other Chromium browsers or on mobile yet, so a team on Edge or Brave cannot use it without switching to Google Chrome.
  • The risk is real and it is managed rather than removed. A safety classifier validates each action before it runs, and probes screen page content for prompt injections (Anthropic). A prompt injection is text hidden on a page that tries to redirect the model. The setting that matters most is auto-approval: it is on by default, and turning it off in settings makes every action wait for your click.
  • Not on your first rollout. Claude can automatically approve actions it judges safe, and you can disable that in settings so each action needs manual approval (Anthropic). Start in manual mode on low-stakes work, watch what it actually clicks, and only widen the permissions once you have seen it handle your real pages.
  • No. Claude in Chrome is generally available on every paid Claude plan, which means the extension carries no separate charge but does require a paid plan (Anthropic). Anthropic publishes current plan tiers on its pricing page, and plan contents change, so confirm there rather than relying on a figure quoted anywhere else.
  • A connector talks to a system's API, and Claude in Chrome works the page the way a person does, using the logins you already have (Anthropic). That makes the extension the option for systems with no usable API, such as vendor portals and internal tools. A connector remains the better choice anywhere a real API exists.
  • No. Anthropic states the extension does not run on other Chromium browsers or on mobile yet (Anthropic). Browser work that happens on a phone stays manual for now, so a field team working from a mobile browser gets nothing from this release.

Automate the work that lives in a browser tab

We build and operate automation inside other people's businesses, including the jobs stuck behind a vendor portal with no API. Bring us the five browser tasks eating the most hours and we will tell you which ones are worth automating.

Book a Consultation