Claude Chrome Extension: What It Does and How to Set It Up in 2026

You’ve probably seen the name Claude attached to a Chrome extension and wondered what it actually does once it’s installed. Short version: the Claude Chrome extension, which Anthropic calls Claude in Chrome, turns Claude from a chat window into something that can read a live webpage and act on it (clicking buttons, filling in forms, navigating between pages) while you watch and decide what it’s allowed to do next.

This isn’t a general AI chatbot bolted onto your toolbar. It’s an agent that operates inside the tabs you already have open, which raises real questions about access, safety, and what happens to your browsing session while it’s working. Below is what the extension does, who can install it, how the permission system works, and where the built-in safeguards stop short — plus a smaller but practical concern: what an agent clicking through your tabs can do to the window you were working in before you started.

What Is the Claude Chrome Extension?

The Claude Chrome extension is a browser extension that allows Claude to read, click, and navigate websites alongside you. That’s Anthropic’s own definition, and it’s a useful one because it draws a clear line: this isn’t Claude summarizing a page you paste in, it’s Claude operating inside the page itself, from a side panel you open in Chrome.

Anthropic’s product page describes the mechanism plainly: Claude “reads the page you’re signed in to, then clicks, types, and fills forms while you decide what happens next.” In practice that means you open the side panel, type a request in plain language (something like “fill in this expense form using the receipt in the other tab”) and Claude works through the page the way you would, one action at a time, with you able to step in.

It’s worth being precise about the name, because search results use several variants. You’ll see it called Claude in Chrome, the Claude browser extension, the Claude extension for Chrome, or the Claude AI Chrome extension. They’re all the same product; Anthropic’s own listing and support articles default to “Claude in Chrome,” and that’s the name used through the rest of this guide.

The extension is published directly by Anthropic on the Chrome Web Store, listed under Extensions and Tools. Its own store description repeats the same framing: open the side panel, describe the task, and Claude reads, clicks, types, navigates, and fills forms while you watch. That data-handling disclosure on the listing also states plainly that the extension handles personally identifiable information, personal communications, location, web history, user activity, and website content — which is the trade-off for an agent that needs to see a page to act on it.

It reads what you can see. Claude in Chrome operates on the page as it’s signed in and rendered for you. It isn’t scraping a separate database of your accounts. If you’re not logged in to a site, it can’t act on it either.

Quick Answer: Claude in Chrome at a Glance

Here’s the short version for anyone who wants the facts before the detail. Availability, browser support, and plan requirements have moved fast since the extension’s pilot, so treat this table as a September 2026 snapshot rather than a permanent state.

If you’re on the free plan and want the functionality, the only path is upgrading to Pro or above. There’s no separate free tier of the extension. And if you’re on Edge or another Chromium browser hoping to find a version of the Claude extension for Chrome that also runs there, Anthropic’s own support documentation rules that out directly.

Which Claude Plans Include the Chrome Extension?

Pro, Max, Team, and Enterprise plans include Claude in Chrome; the free plan does not. That’s stated directly in Anthropic’s getting-started support article, which lists exactly those four plans as having access.

The rollout got there in stages. Anthropic piloted the extension starting August 25, 2025, then after roughly three months of testing opened it in beta to all Max subscribers on November 24, 2025. A follow-up update on December 18, 2025 extended availability to Pro, Team, and Enterprise plans. So if you tried searching for a claude for chrome extension late in 2025 and came up empty on your plan, that’s likely why — the wider rollout came afterward.

For teams, there’s a layer worth knowing before you assume everyone on the account has it turned on the same way:

• Team plans: the extension is enabled by default for the organization, and admins can turn it off if they want tighter control from day one.
• Enterprise plans: the extension was disabled by default. Starting September 10, 2026, it switched to on by default — unless an admin had already disabled it before that date, in which case it stays off.
• Admin controls: both plans manage this from Organization settings > Claude in Chrome, where Claude in Chrome carries its own permission separate from Claude Cowork, so a company can grant one without the other.

Admins on Team and Enterprise plans also get an allowlist and a blocklist. The allowlist specifies which sites Claude is permitted to access, and Anthropic’s own guidance recommends starting with a restrictive one, especially during an initial rollout. The blocklist specifies sites Claude should never access, regardless of any other setting — useful for keeping an agent away from internal tools that were never meant to be automated.

How Do You Install the Claude Chrome Extension?

You install it from the Chrome Web Store, sign in with a qualifying Claude account, and grant the permissions Chrome asks for. Anthropic’s support article lays out the exact sequence, and there’s nothing unusual about it compared to installing any other extension — the difference comes after install, in the permission mode you choose.

Once it’s installed, the side panel is your main workspace. On Max and Team plans, and rolling out to Pro, that panel can run as a Claude Cowork session, which keeps saved conversations and gives you access across devices rather than starting fresh every time you open it.

If you’re not sure which browser you’re currently running, or whether it’s Chrome versus a Chromium-based lookalike, a quick check with a tool like our browser identifier at /tools/what-is-my-browser/ settles it before you go hunting for an extension that won’t install.

What Can Claude in Chrome Actually Do?

Claude in Chrome reads the page you’re on and then acts on it (clicking, typing, navigating, and filling in forms) based on what you type into its side panel. Anthropic’s own store listing describes this as the core loop: you describe what you’re trying to do, Claude reads the page, and it works through the steps while you watch and decide what happens next.

That’s a meaningfully different pattern from a chatbot that only talks back. Where a chat window can suggest text for a form, Claude in Chrome can put that text into the actual fields, move to the next page, and keep going — a multi-step task rather than a single answer. That’s also why it needs a permission system at all: acting on a live page carries more risk than generating a paragraph of text.

Some of what this looks like in practice, based on Anthropic’s own framing of the product:

• Reading a page you’re signed in to and summarizing or acting on what it finds.
• Clicking through a multi-step flow (a form, a settings panel, a checkout page short of the final purchase) on your instruction.
• Typing into fields and submitting forms you’ve described what to fill in.
• Navigating between pages or tabs as part of completing a task.

None of this happens silently by default. The permission mode you’ve selected decides how much Claude checks with you before each step, which is the subject of the next section — and it’s worth understanding well before you hand it a real task.

How Do the Permission Modes Work?

Claude in Chrome has three permission modes, chosen from a dropdown on the chat input, and they control how much Claude checks with you before acting. Anthropic’s permissions guide describes each one in plain terms, and picking the right one for a given task matters more than most people expect.

Site access is a separate layer from the approval mode. You can grant access one action at a time with “Allow this action,” or hand over standing access with “Always allow actions on this site” — and Anthropic’s guide notes you can review and revoke any “always allow” site later from the extension’s Permissions settings page. That’s a good habit to build early, since standing access is easy to grant in the moment and easy to forget about later.

There’s also a hard floor that applies no matter which mode you’ve picked. Regardless of permission mode, Claude requires explicit approval before downloading a file, entering potentially sensitive information, or granting any authorization. So even on Skip all approvals, those three categories still stop and ask.

Press Ctrl + Shift + S (Cmd + Shift + S on Mac) to save a BelayTab session in a couple of seconds, before you open the side panel and describe your task.

What Can Claude in Chrome Not Do?

Claude in Chrome cannot make purchases, create accounts, handle payment or ID data, permanently delete anything, or modify system files — and this holds true no matter which permission mode is active. Anthropic’s permissions guide is explicit that these are actions Claude cannot perform under any circumstances, not just ones it asks about first.

The safety guide adds a few more hard stops on top of that list. Claude in Chrome cannot bypass captchas, cannot trade stocks or make investments, and cannot scrape or gather facial images. These aren’t settings you can toggle on with Skip all approvals. They’re built-in limits regardless of how permissive your setup is.

It’s also worth knowing what the extension simply isn’t designed to do, separate from what it’s blocked from doing. It doesn’t work outside Chrome on desktop — not in Edge, Brave, Opera, or Vivaldi, and not on any mobile browser. It’s also not a general-purpose scraper of your accounts; it only acts on the page currently in front of it, in the session you’re signed in to.

None of this makes the extension weak. It makes it deliberately narrow. An agent that can click through a live page is powerful enough that the boundaries matter as much as the capabilities, which is exactly why Anthropic documents them this specifically rather than leaving them implied.

Is the Claude Chrome Extension Safe to Use?

Claude in Chrome is safer than an unguarded browsing agent would be, but Anthropic itself still calls it risky and asks you to read its safety guidance before use. The core risk the company names directly is prompt injection: malicious instructions hidden in web content (on a website, in an email, inside a document) could try to trick Claude into taking actions you didn’t intend.

Anthropic’s stated defense against that is layered. Claude is trained with reinforcement learning to recognize and refuse malicious instructions, even when they appear authoritative or urgent, and two classifiers sit on top of that — one screening incoming content, another reviewing every action before it executes. The company reports this combination brought the attack-success rate in internal testing that combines known effective techniques under 0.08 percent. A separate figure from Anthropic’s blog post describes the improvement differently: overall attack success dropped from 23.6 percent to 11.2 percent, and browser-specific attack success dropped from 35.7 percent to 0 percent after mitigations.

Those numbers are meaningful progress, not a zero-risk guarantee — which is exactly the tone Anthropic’s own materials take. Official guidance recommends avoiding Claude in Chrome on pages involving financial accounts, medical information, legal documents, or work accounts with sensitive company data, while noting the extension does ask for permission before accessing financial sites specifically. And a line worth remembering before any agentic task: you remain responsible for all browser actions taken by Claude on your behalf, including anything published, any transaction, or any data modification.

If you want a broader look at how Claude in Chrome compares to full AI browsers rather than an extension bolted onto Chrome, our rundown of the best AI browsers in 2026 covers products built around agentic browsing from the ground up. And if you’re building out a general checklist for what you allow to run in your browser at all, the settings walkthrough in how to secure Chrome browser is a reasonable companion piece to this one.

Is There a Separate Claude Code Chrome Extension?

No. There is no separate Claude Code Chrome extension; Claude Code is a launch point for Claude in Chrome, not a second product. Anthropic’s own documentation states that Claude in Chrome’s side panel can be launched from the Chrome side panel itself, from a Claude Cowork session, or from Claude Code, which explains why searches for a “claude code chrome extension” turn up results about Claude in Chrome instead of something distinct.

If you already use Claude Code as a coding agent, this is the connection worth understanding: it’s one of three doors into the same Claude in Chrome experience, not a fork of it with different capabilities. Once the side panel is open, the permission modes, the plan requirements, and the safety limits described earlier all apply the same way regardless of which door you came through.

This distinction matters most for developers who assumed a code-focused agent would come with code-focused permissions — say, broader access to run scripts or modify files. It doesn’t. Modifying system files is on the same blocked list as making purchases, no matter whether you opened the side panel from the toolbar icon or from within Claude Code. The underlying extension, its rules, and its limits are one thing wearing three entry points.

If your workflow already mixes Claude Code with a lot of browser-based research or documentation lookups, it’s worth thinking about tab management as its own concern, separate from the AI tooling. A session manager built for that job (see our overview of the best session manager options in 2026) solves a different problem than any AI extension will: keeping the forty tabs from three different projects sorted and recoverable, whether or not an agent ever touches them.

What Should You Do With Your Tabs Before Turning On Claude in Chrome?

Save your window as a named session before you start an agentic task, because Claude in Chrome is built to click and navigate the way you would, and that means your tab layout can end up different from where it started. Nothing in Anthropic’s documentation suggests the extension deletes tabs outright — but multi-step navigation, opening reference pages, and following links as part of a task can easily leave you with more tabs, fewer tabs, or tabs in a different order than when you handed off control.

That’s a fundamentally different risk than the security concerns covered above. It’s not about data exposure. It’s about losing track of exactly which fifteen tabs made up the research project you were in the middle of, or the flight-booking window with three comparison tabs still open next to it, before you asked Claude to fill in a form on one of them. Browsers don’t keep a clean undo button for “what my window looked like ten minutes ago.”

This is where a purpose-built extension for tabs, rather than pages, earns its place alongside an AI agent rather than in competition with it. BelayTab saves an entire window (tab groups with their names and colors, pinned tabs, and tab order) as a named session with one click or Ctrl/Cmd+Shift+S, and it works whether you triggered the change yourself or an agent did. It also takes an automatic snapshot of every open window on a five-minute timer and again when a window closes, so even a session you forgot to save by hand has a recent backup sitting in the Recovery screen.

• Before a long agentic task: save the window as a named session so you have a clean point to return to.
• If a task goes sideways: open the Recovery screen and restore the version from just before you started, without losing the version Claude left behind.
• For recurring workflows: keep separate named sessions for separate projects, so an agent’s mid-task detour never gets tangled with other work in the same window.

None of this replaces Claude’s own permission modes or Anthropic’s safety guidance. Those are still your first line of defense for what Claude is allowed to do. BelayTab covers a narrower, more mechanical gap: making sure the tabs you had before an agent started are still there, in order, no matter what happened while it worked. You can read more about how that recovery layer works on our automatic snapshots page, or see the full picture of saving a window in one action on the save tabs feature page.

If you manage extensions across a whole team, alongside Claude in Chrome, it’s worth reading through our guide to managing Chrome extensions to keep both the AI tooling and your recovery tooling under one clear policy rather than an accumulating pile of one-off installs.

Where Can You Verify These Facts Yourself?

Anthropic’s own support articles are the most current source, and they’re worth bookmarking rather than relying on a single guide to stay accurate. AI product details, permission defaults, and plan availability all move fast, and this space in particular has changed more than once in the past year alone.

Three specific pages are worth keeping close if you’re using Claude in Chrome regularly: the getting-started article for install steps and plan availability, the permissions guide for exactly what each mode does, and the safety article for the prompt-injection defenses and sites to avoid. Anthropic updates these directly rather than through blog posts, so they’re the first place a genuine change in behavior or eligibility would show up.

If you install extensions through the Chrome Web Store regularly and want a general framework for vetting any listing (not just this one) our guide on how to find, install, and vet extensions through the Chrome Web Store walks through the habits worth building regardless of which extension you’re looking at. The same due diligence that applies to an AI agent with page access applies, in a smaller way, to anything you give tab or bookmark permissions to.

Circling back to the core question this guide opened with: the Claude Chrome extension is a real, actively maintained product from Anthropic, gated to paid plans, limited to desktop Chrome, and built with a permission system precisely because it can act on pages rather than just talk about them. Read the official guidance before you turn it loose on anything sensitive, and treat your own tab layout as a small, solvable side problem rather than something you have to manage by memory.