It reads the 2FA code from Gmail

The problem this solves

An agent automating a real task hits a login that sends a one-time code to your email. A headless browser is a stranger here: it isn't signed into your Gmail, so the flow dead-ends. The usual workarounds - email-API integrations, forwarding rules, manual copy-paste - are exactly the friction you were trying to remove.

The walkthrough

1. The agent hits the 2FA prompt. It's filling a login on some site (using browser_fill and browser_click) and the site says "we've emailed you a code."
2. It opens your Gmail - already signed in. browser_navigate to mail.google.com with new_tab: true. There's no login step, because this is the Chrome where you're already authenticated. Your session, your cookies.
3. It reads the newest message. browser_get_page_content returns the inbox text; the agent finds the latest sender/subject that matches and reads the code out of the body.
4. It types the code back and continues. browser_switch_tab back to the login tab, browser_fill the code, submit, done - the task carries on past the wall.
5. If it's unsure, it asks you. browser_ask_user lets the agent pause and confirm with a human before doing anything sensitive - a built-in checkpoint, not an afterthought.

That's the whole move. No new integration, no secret to store. The code goes back to your agent like any tool result, and is never sent to us.

What it looks like

You:     Log into the billing portal and download this month's statement.

Claude:  [browser_navigate, browser_fill, browser_click]
         The site says it has emailed a code.

         [browser_navigate  mail.google.com, new_tab: true]
         [browser_get_page_content]
         The newest message is from billing@ - the code is 418293.

         [browser_switch_tab back to the portal]
         [browser_fill, browser_click]
         Signed in. statement-2026-08.pdf is in your Downloads.

Five tools, no new integration, and nothing typed by you. The Gmail tab needed no login because it is the Chrome you were already signed into.

Why most tools can't do this

Playwright, Puppeteer and other headless setups spawn a fresh browser with none of your sessions. To read a 2FA email they'd need you to hand over Gmail credentials or wire up the Gmail API - which is the friction, not the fix.

Playwright MCP's Chrome extension is the honest exception to the first half: it reaches the Gmail tab you are already signed into. What it has no answer for is the second half. Its 72 tools contain nothing that can stop mid-run and ask you for a code (checked 2026-09-19), so a step that genuinely needs a human ends the run rather than pausing it. Browser MCP sidesteps both: it is never a stranger to your accounts, and browser_ask_user can hand the step to you and carry on in the same tab.

One honest boundary

This works on your own accounts, that you're already logged into, for things you're authorized to do - the same boundary as everything else Browser MCP does. It is not a way around 2FA on accounts that aren't yours; the code still lands in an inbox only you control. See When NOT to use Browser MCP for where that line sits.

FAQ

Does the 2FA code leave my machine? Not to us. The agent reads it in your own browser and types it back. Like anything the agent reads, it passes through your AI client and that client's model provider - but it is never sent to a server of ours, because there isn't one.

Do I need the Gmail API or an app password? No. The whole point is that it uses the Gmail you're already signed into in Chrome - no API, no key, no forwarding rule.

Can I keep the agent from acting without me? Yes - the browser_ask_user tool lets it pause for a human decision at any step you care about.

Does it work with Outlook / other webmail? Yes - any webmail you're logged into in your Chrome. Gmail is just the common case; the mechanism is "open your signed-in mail in a tab of its own and read it."

How do I set this up? Install Browser MCP for Claude Code, then ask your agent to complete a flow that hits a 2FA wall.