The Claude Chrome extension, explained

People searching for a Claude Chrome extension usually mean one of two things: Anthropic’s official Claude in Chrome, which is an agent that operates your browser for you, or simply having Claude’s models available while you browse, for summarizing, asking, and drafting. Those are very different tools. This guide covers the official one first, including the parts Anthropic itself tells you to be careful about, then the second option.

What Claude in Chrome actually is

Claude in Chrome is Anthropic’s own extension, and it’s not a chat sidebar. It’s an agent: give it a task and it navigates, clicks buttons, types into fields, fills forms, and moves between tabs to get it done. Think “collect the numbers from these three dashboards into a sheet” or “unsubscribe me from everything in this inbox that looks like marketing”, not “explain this paragraph”. It can also run tasks in the background and complete scheduled workflows.

As of mid 2026 it’s a beta feature, and Anthropic is unusually direct that beta here means real risks, not rough edges. More on that below.

What you need before installing

• A paid Claude plan. It’s included with Pro, Max, Team, and Enterprise. There’s no free tier for the extension itself.
• Actual Google Chrome. Anthropic’s docs are explicit that other Chromium browsers like Brave and Edge aren’t supported, and neither is mobile.
• Model choice comes with it. Per Anthropic’s documentation it works with all public Claude models, so you’re not locked to a lesser model by plan.

Installation is ordinary: get it from the Chrome Web Store, sign in with your claude.ai account, and grant it access. One genuinely important tip: the store is full of third-party extensions with “Claude” in the name, some of which resell API access at a markup. Check that the publisher is Anthropic before you install anything.

The part to understand before you turn it on: prompt injection

An agent that reads pages and acts on them can be steered by pages. Hidden text in a website, an email, or a document can contain instructions that the model mistakes for yours, and a hijacked agent has your logged-in sessions: your email, your accounts, whatever Chrome can reach. This is called prompt injection, it is the central risk of every browser agent, and Anthropic names it openly in its own safety notes rather than burying it.

The extension ships with guardrails: you pre-approve which sites Claude may act on, you can review its plan before letting it run, and it’s designed to ask before high-risk actions like purchases. Security researchers keep finding edge cases anyway, which is what beta means here. The practical posture: start it on low-stakes sites, watch the first runs, and keep it away from banking and anything you can’t undo. If you already pay for Claude and want web chores delegated, it’s the real thing and it’s genuinely capable. Just extend it trust the way you would a new assistant, gradually.

If you want Claude while you browse, without the agent part

A lot of “claude chrome extension” searches aren’t looking for delegation at all. They want to read a page and ask Claude about it, summarize a long article, or draft a reply, with a model they choose. You don’t need an agent for that, and there’s an argument you don’t want one: a tool that can only read can’t be tricked into clicking.

That’s the job our extension, UtilEngine, does. Press your shortcut (Alt+Z by default) on any page for Ask AI and Chat: summarize, question, rewrite, with the page as context. It’s bring-your-own-key: Claude access comes through your own OpenRouter key, one key that covers Anthropic’s models and everyone else’s, priced per token instead of per month. The model list has 67 entries including twelve Claude models, switchable per conversation. Its AI tools (web search, searching your open tabs, history, and bookmarks) are strictly read-only and show up as visible chips in the answer: it never clicks, types, or acts on a page for you, by design. And it runs in Chrome, Brave, and Edge, since it doesn’t need Chrome-only agent plumbing.

The honest caveat, both directions: UtilEngine is paid ($49 once, plus your token costs) and it will not do your web chores for you. If delegation is what you came for, install Anthropic’s extension. If an assistant that reads alongside you is what you came for, that’s us, and the two coexist fine in the same browser.

Common questions

Is there a free Claude extension for Chrome?

Not from Anthropic; the official extension requires a paid plan. The free way to use Claude in a browser is claude.ai itself in a tab, which has a free tier. Be wary of unofficial free “Claude” extensions, since the model access has to be paid for somehow, and it’s usually your data or a marked-up subscription.

Does Claude in Chrome work in Brave, Edge, or Arc?

No. Anthropic’s documentation says it’s unsupported outside Google Chrome itself, and mobile browsers are out entirely. That’s worth knowing if you’ve deliberately left Chrome; BYOK tools like ours cover the assistant use case in Brave and Edge, but the official agent needs Chrome.

Which Claude model does the extension use?

Anthropic’s docs say all public Claude models are available in it, so the current generation of Claude models is selectable rather than fixed.

Is Claude in Chrome safe to use?

Anthropic labels it a beta with unique risks, which is the right way to hear it. The safeguards (site allowlists, plan review, confirmation before irreversible actions) are real, and so is the underlying prompt injection problem for all browser agents. Use it where mistakes are cheap; earn up to the important stuff.

Do I still need Claude in Chrome if I have UtilEngine, or vice versa?

They don’t overlap much. UtilEngine gives you Claude (and 55 other models) for reading, asking, and writing, paid per token, and never acts on pages. Claude in Chrome acts on pages, needs a Claude subscription, and is the only official way to have Claude do things in your browser. Plenty of people will reasonably run both.