It is a browser extension that lets Claude see and work with the websites you are using.

According to Anthropic’s Help Center, Claude in Chrome is a browser extension that allows Claude to read, click, and navigate websites on your behalf. After you install it, a Claude icon appears in your Chrome toolbar. Clicking it opens Claude in a side panel that stays visible while you browse, so you can chat with Claude about the page in front of you without switching tabs or windows.

The key difference from the normal Claude chat is that Claude is not limited to what you paste into the message box. In the extension, Claude can look at the page you have open, and, with your permission, take actions on it: scrolling, clicking buttons, filling in fields, and moving between pages. This is why it is called a browser-using AI helper rather than just a chat window.

flowchart LR
    A["You browse a website in Chrome"] --> B["Open the Claude side panel"]
    B --> C["Ask Claude for help with the page"]
    C --> D["Claude reads the page and may act with your permission"]
    D --> E["You review the result"]
        

A few simple words appear again and again in this topic. Here they are in plain English.

Availability depends on your plan and your browser.

According to the Help Center, Claude in Chrome is available in beta for all paid plans, which are Pro, Max, Team, and Enterprise, on the Chrome web browser. It is not supported on other Chromium-based browsers or on mobile devices, so it works in Google Chrome on a computer. Plan names and availability can change, and organizations can control whether staff may use it, so check the current Help Center article for your situation.

Setting it up takes only a few minutes.

It can help with reading, comparing, and doing repetitive steps in your browser.

Understanding what Claude can see helps you decide when to use it.

Anthropic’s safety guide explains that when you open the Claude side panel, Claude takes screenshots of your active browser tab to understand the content of the page. In other words, whatever is visible in that tab can be seen by Claude. The guide adds that Claude cannot filter sensitive content out of what it sees. That is why Anthropic recommends being mindful of what is visible and avoiding opening the extension while you are viewing sensitive information or documents.

A layered permission system decides what Claude may do.

The Help Center’s permissions guide explains that Claude in Chrome uses a multi-layered permission system. In the side panel, a menu near the chat input lets you choose a permission mode. Two modes are described. In the manual mode, Claude pauses and asks for approval before each action, and you choose to allow or deny. In the automatic mode, Claude keeps working and reviews each action for safety, automatically blocking anything it judges unsafe and pausing to ask you when needed. Names of modes and defaults can differ by version, so check the guide for what your screen shows.

Some actions always need your approval

Even if you choose to always allow actions on a site, the permissions guide says Claude still asks for your explicit approval before certain high-risk steps. These include downloading a file, entering potentially sensitive information into a page, granting authorizations, making purchases or financial transactions, permanently deleting files or data, modifying permission settings, and creating accounts.

flowchart TD
    A["Claude wants to act on a site"] --> B{"Is it a high-risk step?"}
    B -- "Yes: purchase, download, delete, sensitive data" --> C["Claude asks for your approval every time"]
    B -- "No" --> D{"Which permission applies?"}
    D -- "Manual mode or allow once" --> E["Claude asks first"]
    D -- "Always allowed for this site, or automatic mode" --> F["Claude proceeds, screening each action"]
        

You can review and change which sites Claude may use at any time.

The extension includes settings where you can see which sites have “always allow” status, remove permission for specific websites, and look at your permission history. The Help Center explains that you reach these from the extension’s settings, using the menu in the side panel. If you ever feel unsure about a site, remove its permission and let Claude ask each time.

Sites Claude will not use, or asks about first

For safety, Anthropic states that Claude cannot access certain high-risk sites, such as adult content websites and known pirated content sites. It also states that Claude asks for permission before accessing financial sites. These lists and rules may be updated, so the safety guide is the place to confirm the current details.

The biggest danger for any AI that browses is hidden instructions inside web content.

Anthropic’s safety article explains that the biggest risk facing browser-using AI tools is prompt injection. In a prompt injection attack, malicious instructions are hidden in web content, such as websites, emails, or documents, and they try to trick the AI into taking actions you never asked for. The text might be invisible to you, for example white text on a white background, but visible to the AI that reads the page.

Several layers of protection work together, though none is perfect on its own.

A few simple habits make a big difference.

Never hand your passwords to an AI. There are safer ways to sign in.

Many useful tasks need you to be signed in to a site. The safest general rule is that you should enter passwords and one-time codes yourself, or use a trusted password manager, instead of typing them into a chat. Anthropic notes that with its 1Password integration for Claude, Claude can complete tasks that require signing in without handling the credential itself. The password manager fills the login directly, and passwords and one-time codes never enter Claude’s view.

Also keep in mind that Claude asks for your explicit approval before entering potentially sensitive information into a page. When that prompt appears, stop and think. Ask yourself whether this task really needs that information, and whether you trust the site.

Companies can decide whether and how their staff use the extension.

On Team and Enterprise plans, administrators have controls. According to the Help Center, an owner can switch Claude in Chrome on or off for the whole organization. On Team plans the extension is enabled by default, while on Enterprise plans it is disabled by default until an owner turns it on. Admins can also use allowlists to restrict Claude to approved sites, and blocklists to keep it away from certain sites no matter what a user permits.

Anthropic recommends starting with a restrictive allowlist for the security of organizational data, then expanding access gradually as the organization becomes comfortable. For organizations covered by HIPAA, the safety article says Claude in Chrome is off by default and is not covered by the Business Associate Agreement, and it recommends against using the extension on pages with regulated data.

Let us follow an imagined college student named Dev as he uses the extension carefully.

Dev used the extension well. He started with a simple, low-risk task, gave a precise request with clear limits, verified the results, and kept important decisions in his own hands.

Because Claude may act on real websites, your wording matters even more than in an ordinary chat.

Most problems come from a few risky habits.

A browser helper saves effort, with real risks to respect.

Keep these habits in mind whenever you use Claude in Chrome.

Quick answers to the questions beginners ask most often.

A browser extension that lets Claude read, click, and navigate websites on your behalf, shown in a side panel that stays visible while you browse.

Google Chrome on a computer. According to the Help Center, other Chromium-based browsers and mobile devices are not supported.

The Help Center describes it as available in beta for paid plans: Pro, Max, Team, and Enterprise. Check the current article, since availability can change.

When the side panel is open, Claude takes screenshots of your active browser tab to understand the page. It cannot filter out sensitive content, so avoid opening it on pages with confidential information.

That depends on the permission mode. In manual mode it asks before each action. In automatic mode it works on its own with safety screening. In every mode, certain high-risk steps, such as purchases, downloads, and permanent deletions, still need your approval.

Hidden instructions placed in web content that try to trick an AI into doing something you did not ask for. It is the main risk for browser-using AI, and safeguards reduce it but cannot remove it entirely.

No. Anthropic strongly advises against using it to manage financial accounts or investments, or to handle legal documents or contracts, and recommends against using it on pages with regulated data.

Use a separate browser profile, start with simple tasks, use manual approval on new sites, write specific prompts with limits, and stop immediately if anything looks odd.

Yes. On Team and Enterprise plans, administrators can turn the extension on or off for the organization and use allowlists and blocklists to control which sites Claude can use.

Sign in yourself, or use a supported password manager so that passwords never enter Claude’s view. Do not paste passwords or codes into the chat.

Yes. In the extension settings you can review sites with “always allow” status, revoke permissions, and see your permission history.

In the Claude Help Center articles “Get started with Claude in Chrome”, “Use Claude in Chrome safely”, and “Claude in Chrome permissions guide”, with a separate article for admin controls.

Claude in Chrome is a browser extension that brings Claude into a side panel beside the page you are viewing, where it can read, click, and navigate websites on your behalf. It is available in beta for paid plans, works in Google Chrome on computers, and is useful for summarizing pages, comparing information, filling forms, and handling repetitive browsing chores. You stay in control through permission modes and site permissions, and some high-risk steps always need your approval. The main danger is prompt injection, where hidden instructions in web content try to steer the AI, and although classifiers, screening, and blocklists reduce the risk, they cannot remove it. Safe use means a separate browser profile, simple tasks to begin with, manual approval on new sites, specific prompts with clear limits, no sensitive financial or legal tasks, and stopping at the first sign of odd behavior.