Browser MCP vs Playwright MCP

One difference is left, and it is the only one we will defend: their 72 tools contain nothing that can stop mid-run and ask you for something. No 2FA code, no CAPTCHA hand-off, no "which of these three accounts did you mean". That is not an oversight - Playwright MCP grew out of a testing tool, and a test that asks for help is a failed test. Browser MCP is built the other way round: browser_ask_user pauses, asks you on your own screen, and continues in the Chrome you're signed into.

The architectural difference (everything else follows from it)

Playwright MCP has two modes, and the difference between this page's two products lives in the first one. By default it launches and manages its own browser: an isolated in-memory profile (--isolated) or a persistent-but-dedicated one. That design is what makes it reproducible in CI and safe to parallelize, and it is why such a profile starts out logged in to nothing of yours. Since mid-2025 it has also shipped a Chrome extension that connects to tabs in the Chrome you already use, with your logins, and supports several clients at once, each in its own coloured tab group. In that mode the logged-in browser is no longer a difference between us at all.

Browser MCP is a Chrome extension plus a local stdio server. There is no second browser: tools act on the Chrome window you already use, through the Chrome Debugger API. Nothing to re-authenticate, because it is your authenticated browser. The trade-off is symmetrical: you get exactly one browser - yours.

(We publish the download gap on purpose. Playwright MCP is the bigger project by three orders of magnitude; this page is about when that is not the deciding axis.)

The same task, side by side

You:      Sign in to the admin dashboard and export last month's invoices.

Playwright MCP:  [launches its own browser]
                 I'm at the login page. I need credentials.

Browser MCP:     [browser_navigate]
                 You're already signed in. Exporting now.

Against Playwright MCP's default mode, that is the difference: the second exchange has no login step because the tab is in the Chrome you use. In its extension mode, both start signed in.

We measured the reverse case too, on 19 September 2026, and reported a miss on a framework-controlled <select>. The fault was our own test page, not React: 1.30.0 lands the choice on real React. The measurement and the retraction are on /learn/tools-that-lie/, because a comparison page that only lists what we win is not a comparison.

Can Playwright MCP use my real browser too?

Partly, and it is fair to say so: Playwright MCP has an opt-in --extension flag - "Connect to a running browser instance (Edge/Chrome only). Requires the 'Playwright Extension'" (its README, checked 2026-07-21). It is not the default, not the primary documented path, and the managed profile remains the designed-for mode. If real-Chrome sessions are the core of your workflow rather than an edge case, a tool where that is the only mode has fewer seams.

When Playwright MCP is the right choice

• CI/CD pipelines and scheduled jobs - headless, reproducible, no human's browser involved
• Test suites and scraping that benefit from parallel isolated instances
• Anything on infrastructure (serverless, containers) rather than a workstation
• When you want the largest tool surface (72 tools incl. PDF, vision, network interception)

We list these plainly in When NOT to use Browser MCP - they are the same list.

When Browser MCP is the right choice

• The workflow starts behind a login you already have - dashboards, admin panels, webmail
• 2FA-gated sites: the agent can wait while you approve, or open your signed-in mail in a tab of its own and read the code
• Sites where fresh headless profiles get blocked or endlessly challenged
• "Do this in my browser" tasks: triage my tabs, fill this form as me, pull a token from my dashboard

FAQ

Is Browser MCP a Playwright wrapper? No. It is a Chrome extension using the Chrome Debugger API on your running browser; no Playwright, no bundled browser binary.

Can I run both? Yes - they register as separate MCP servers and many setups use Playwright MCP for CI and Browser MCP for logged-in interactive work.

Which is safer for credentials? Both run locally. Browser MCP does not need your passwords for sites you are already signed into. If a step does need one, browser_ask_user asks you on your own screen, and what you type goes to your AI client like any other tool result. It operates a browser where you are already signed in, and your cookies stay in that browser unless your agent explicitly reads them.

Is Playwright MCP really maintained by Microsoft? Yes - the repository lives under the microsoft GitHub organization (checked 2026-07-21).