Skip to main content

Claude Guard for Claude in Chrome

Claude Guard is a lightweight Chrome extension that acts as a request guardian while a Claude browser agent ("Claude in Chrome") is active on a page. It monitors, blocks, pauses, and gates outgoing HTTP requests behind human approval, so risky automated actions stay visible and under your control.

Download and install​

Install Claude Guard from the Chrome Web Store:

Claude Guard on the Chrome Web Store

  1. Open the Chrome Web Store listing.
  2. Click Add to Chrome, then confirm with Add extension.
  3. (Optional) Pin the extension: click the puzzle-piece Extensions icon in the toolbar and pin Claude Guard so its icon is always visible.

That's it — no build step or configuration is required to start using it.

How it works​

When Claude Guard detects an active Claude agent on a page (via the agent's on-page glow indicator), it intercepts outgoing requests and applies your rules:

  • Auto-mode blocks non-GET/HEAD/OPTIONS requests to the same site, and any request carrying an Authorization header, unless you've explicitly allowed it.
  • Blocked requests open a human approval popup where you can allow once, always allow (by hostname), or deny.
  • Denied requests get a synthetic 403 response so the agent can handle the rejection gracefully.
  • Intercepted APIs include fetch, XMLHttpRequest, form submissions, and navigator.sendBeacon.

How to use it​

Toolbar popup​

Click the Claude Guard icon in the toolbar to open the popup, where you can:

  • Toggle Auto mode on or off.
  • Add and manage allow / block rules (glob patterns supported, e.g. *://api.github.com/*).
  • View recent request stats and logs.

Approving or denying a request​

When Claude Guard holds a request, an approval window appears with the request's method, URL, headers, and body preview. Choose:

  • Allow once — permit just this request.
  • Always allow — permit future requests to the same hostname.
  • Deny — reject the request (the agent receives a 403).

Knowing when it's active​

While a Claude agent is active, Claude Guard shows visual indicators: a tab badge, an on-page banner, toasts, and a title update. When no agent is present, the extension stays out of the way.

Deploy to managed Chrome with MDM​

Admins can force-install Claude Guard and push allow/block rules to managed Chrome browsers through a Chrome configuration profile. Rules delivered this way are marked as managed and cannot be edited or removed by users (they show an MDM tag in the extension popup).

Willow generates the profile for you — no need to hand-write policy.

  1. In Willow, open Security → Claude Guard.
  2. Click Download claude-guard-chrome.mobileconfig. The profile force-installs the extension and ships example allowList / blockList rules.
  3. Open the downloaded profile in a text editor and edit the rules to fit your organization. Each rule has a pattern (a glob such as *://api.github.com/*) and a method (an HTTP method, or * for any):
    • allowList — requests to always permit while a Claude agent is active.
    • blockList — requests to always deny.
  4. Upload the edited profile to your MDM (Jamf, Iru (formerly Kandji), Mosyle, Intune, JumpCloud, etc.) as a custom configuration profile and assign it to your device group.
Policy is read at launch

After the profile applies, fully quit and reopen Chrome (Cmd-Q on macOS) so it re-reads managed policy.

Troubleshooting​

SymptomCheck
Extension doesn't installConfirm the profile applies to the browser and that Chrome is managed.
Requests aren't being gatedConfirm a Claude agent is active on the page and Auto mode is on.
Managed rules don't appearReload policy at chrome://policy, then fully quit and reopen Chrome.

See also: General Troubleshooting

On this page