Claude in Chrome Tutorial: 7 Browser Workflows

See the workflow in action, tap through the tabs below:

claude --chrome

# then, inside the session:
/chrome        # check status, reconnect, pick a browser

Open localhost:3000/dashboard, reload it, and check the console
for errors matching "TypeError|401|CORS". Tell me which file
probably causes each one, then propose a fix. Do not edit anything yet.

Open the settings page of our staging admin console. READ ONLY.
List every setting that looks risky (public access, weak session
timeouts, disabled MFA prompts). Do not click Save or change anything.
Give me a table: setting, current value, why it matters.

What Claude in Chrome actually is

Claude in Chrome is a browser extension that lets Claude see, click, type, and navigate inside your real Chrome window, using the sessions you are already logged into. It works on all paid plans: Pro, Max, Team, and Enterprise. The part DevOps and security people should care about: Claude Code can drive it too, so the same agent that edits your repo can open localhost, poke the form, and read the console. No API connector required.

If you have not wired up tools yet, start with our MCP connectors guide. Connectors are for systems with an API. The browser is for everything else.

Quick setup

Install the extension from the Chrome Web Store, sign in with your Claude account, and pin it. For Claude Code you need extension version 1.0.36 or later and a login through /login (API keys and setup-token logins keep Chrome integration off). Then launch with the flag:

claude --chrome

Run /chrome any time to check status, reconnect, or pick a browser. Heads up: it is not supported in WSL. The official getting started guide covers the side panel flow.

The mindset: a fast intern with your cookies

The browser agent inherits your logins. That is the power and the risk. Think of it as a very fast intern who is already signed in as you. You would not hand that intern the production console on day one, so start read-only, one site, one task, and check the evidence before you trust it with writes.

Claude in Chrome has three permission modes: Manually Approve, Automatically Approve, and Skip All Approvals. Automatic mode runs safety checks and pauses when something needs you, but it burns more of your usage. Skip All Approvals turns the checks off. Save it for sandboxes you own.

7 workflows worth stealing

1. Console-first debugging

Filter, do not dump. Console logs are noisy, so tell Claude what patterns to look for.

Open localhost:3000, reload, and read console errors matching
"TypeError|401|CORS". Group by root cause and point me at the file.

2. Test the change you just made

I changed the login validation. Submit the form with an empty
email, a bad email, and a 300 character password. Report which
error messages appear and which do not.

3. Read-only settings audit

Open the staging admin console settings. READ ONLY. List risky
settings (public sharing, long sessions, MFA off) in a table.

4. Evidence capture

Ask for a screenshot saved to disk, or a recorded GIF of a flow for a ticket. Review it first: recordings capture everything visible, including account details.

5. Data extraction

On the status page, extract each service name, current state, and
last incident date. Save as services.csv.

6. Form filling from a file

For each row in contacts.csv, add a contact in our staging CRM.
Stop and ask if any field looks wrong.

7. Shortcuts for the boring stuff

The extension supports saved shortcuts and scheduled tasks. Turn your weekly "check these five dashboards" routine into one. Keep scheduled runs read-only.

Safety and gotchas

Web pages can contain text that tries to boss your agent around. Anthropic's own permissions guide says Claude will not complete instructions found in emails or web content, but you are still the last line of defense. Keep these habits:

Use a dedicated Chrome profile for agent work, logged into as few things as possible. Keep approvals on for downloads, sensitive data, and anything that grants access. Watch the pause points: Claude stops at login pages and CAPTCHAs, and that is your cue to step in, not to hand over a password. Dismiss JavaScript dialogs by hand, since they freeze the browser channel. And if tools go quiet mid-session, the extension's service worker probably went idle, so run /chrome and choose Reconnect extension. Team and Enterprise admins can also set site allowlists and blocklists, which is worth doing before rolling this out to a whole team.

Usage and cost tips

Automatic approval mode uses more of your usage limit than manual, so match the mode to the risk. In Claude Code, enabling Chrome by default loads the browser tools into every session and increases context use, so use --chrome only when you need it. Uploads are capped at 10 MB per call. Finally, ask narrow questions ("errors matching 401") instead of "read everything on this page."

FAQ

Does Claude in Chrome work in other browsers?

The extension is for Google Chrome. Claude Code's integration also works with Microsoft Edge and detects other Chromium browsers such as Brave, Arc, Vivaldi, and Opera.

Can it make purchases or create accounts?

No. Anthropic's permissions guide lists purchases, account creation, and permanent deletions as things it will not do, and it asks before downloads or entering sensitive data.

Is it safe to leave running unattended?

Only for read-only tasks on sites you trust. Anything that writes, deletes, or touches production deserves a human in the loop.

Your next move

Pick one boring, read-only browser chore this week and hand it to the agent with a tight prompt. Want the full playbook for building safe AI workflows on your team? Check out our courses.