Claude in Chrome (2026): What It Does, Which Plans Include It, and the Settings to Change First

Claude in Chrome: the short answer

Every "AI browser agent" demo shows it booking a flight, which is the one job you should never give it. As of September 2026, Claude in Chrome is Anthropic's generally available Chrome extension: it reads the page you are signed in to, then clicks, types, and fills forms while you approve what happens. It comes with every paid Claude plan, from Pro at $20 a month, with no separate price. Use it for the tedious browser work no integration reaches. Keep it away from money and passwords. And if you run a Team or Enterprise account, check two admin settings this week. This page is what it does, who gets it, what it is good for in a small business, and the safety settings, with every plan fact read off Anthropic's product page, pricing page, and help center on September 12, 2026.

What Claude in Chrome actually does

Anthropic's description is plain: Claude in Chrome "reads the page you're signed in to, then clicks, types, and fills forms while you decide what happens next." That is the whole product. The chat version of Claude answers questions about the web. This version operates the web, inside your own logged-in tabs.

The details that matter, from Anthropic's help center: you can drag several tabs into Claude's tab group and it works across all of them at once. It keeps working when you switch to another tab, as long as Chrome stays open, and it can notify you when it needs permission or finishes. In the classic side panel you can record yourself doing a task once and Claude learns to repeat it. You can save a task as a shortcut and schedule it daily, weekly, monthly, or annually. And for developers, it reads the browser console, network requests, and page structure.

The honest framing: it is a very fast temp who can use any website you can, and who will follow instructions it finds on a page if you let it. Hold that second clause in your head for the rest of this page.

Which Claude plans include Claude in Chrome

All paid plans: Pro, Max, Team, and Enterprise. Not Free. Anthropic's pricing page lists no separate charge for the extension, so the cost is the plan you already pay for:

Two limits Anthropic states outright: it works in Google Chrome on desktop only, "not supported on other Chromium-based web browsers or mobile devices," so Arc, Brave, and your phone are out. And it runs on your plan's normal Claude access, so the plan decision is the same one you would make without it. The full plan-by-plan picture is on Claude AI pricing.

Three ways in: side panel, Cowork, Claude Code

Anthropic gives the extension three front doors, and picking the right one saves you a lot of confused clicking:

• The side panel, when the work is on the page in front of you. Click the Claude icon in the toolbar and it sits beside the tab. On Max and Team it now runs as a Cowork session, so your skills and connectors come along.
• Claude Cowork, when the browser is one step in a longer job. Anthropic's example is the good one: Chrome visits competitor sites and pulls pricing, then Cowork turns it into a formatted deck. If Cowork is new to you, start with how to use Claude Cowork, because the browser is only as useful as the task you hand it.
• Claude Code, when you are building the site you are testing. Code writes the change, Chrome clicks through it and reads the console.

One wrinkle from the same help center: Cowork now has its own browser built into the desktop app, with nothing to install. If you already use Claude in Chrome, Chrome stays Cowork's preferred browser. This is the fourth Claude surface that comes with a paid plan, next to the Office add-ins covered in Claude for PowerPoint.

What it is good for in a small business

The sweet spot is the web that has no API. Anthropic lists "internal tools, legacy portals, admin consoles, and invoice systems," and every small business I have worked with has at least three of those: the supplier portal that exports nothing, the insurance carrier's dashboard, the booking system from 2014. Claude works with the logins you already have.

Jobs I would hand it this week, all straight from Anthropic's own examples: pull last month's numbers out of an analytics dashboard into a summary; read your calendar and draft CRM activity logs for each sales call, which you approve before anything is created; visit five competitors' pricing pages and build the comparison; sort a messy Google Drive into folders and flag the duplicates for you to review; and scan an inbox for promotional mail and present the list before deleting. The recurring ones become scheduled shortcuts. The pattern is read, gather, draft, and wait for you. That is where a browser agent earns its keep.

What it should never touch

Anthropic is refreshingly blunt here, which is more than the influencers selling "let AI run your browser" courses will tell you. The product page says hidden instructions on a web page "can try to redirect Claude's actions," and its advice is: "Skip banking, health records, and anything with credentials you wouldn't share." The help center adds financial transactions, password management, and sensitive personal data to the avoid list.

Translate that into small-business terms: no bank portals, no payroll, no payment processors, no customer health or legal records, no admin consoles where one wrong click deletes something. If your team does not have a written rule for which data goes into AI tools at all, the one-page version is in AI acceptable use policy; add a line that says the browser agent follows the same data classes as everything else.

The safety settings to change first

There are four controls, and three of them are yours to set:

• Permissions Mode. Grant access one site at a time, so Claude cannot reach anything you have not approved. This is the default I would start everyone on.
• Auto-approve with guardrails. Optional. Claude works through a task without asking at each step, and a separate safety check reviews each action for risk and hidden instructions. Turn it on only for sites and workflows you already trust.
• Sensitive-action stops. Built in: purchases, financial actions, and other one-way doors wait for you regardless.
• Admin controls (Team and Enterprise). Under Organization settings > Claude in Chrome, admins switch the extension on or off for the whole company and set allowlists and blocklists.

Check the defaults, because they are not what most owners assume. On Team plans the extension is enabled by default. On Enterprise it was disabled by default and, per Anthropic's admin article, "Starting September 10, 2026, it turns on by default unless you've already disabled it," and users are not notified. Anthropic's own recommendation is to start with a restrictive allowlist and widen it as you get comfortable. For a 10-person company that means a list of maybe six sites, not the whole internet.

What to do today

If you are on a paid plan, install the extension in Chrome, pin it, leave it in Permissions Mode, and give it one read-only job: "pull last month's numbers from this dashboard into a summary." If you are the admin on a Team or Enterprise account, open Organization settings > Claude in Chrome before you do anything else and decide whether the default is the setting you actually want. And if you want Claude set up properly for the whole team, with the safe jobs identified, the allowlist written, and the first three workflows running instead of a pile of extensions, that is the Claude for small business install, from $4,500 in about two weeks; the $397 Playbook is the do-it-yourself version.