Writing automated browser tests is slow, and maintaining them is slower. Playwright MCP takes a different route: an AI agent reads your test cases and executes them in a real Chromium browser — clicking, typing, navigating and verifying — while you watch live. When it hits a login or a CAPTCHA, it hands control to you, then carries on. No scripts, no selectors, no CI wiring. Just your test cases, running themselves.
The gap between test cases and test automation
Most QA teams have hundreds of well-written manual test cases and a fraction of them automated. The reason is familiar: turning a test case into a Playwright or Cypress script takes engineering time, the scripts break every time the UI shifts, and someone has to own the framework. So the manual cases get run by hand, sprint after sprint, or not at all.
Playwright MCP closes that gap from the other side. Instead of asking you to translate a test case into code, it reads the test case the way a human tester would — preconditions, steps, expected result — and performs it in a live browser. The test case you already wrote is the automation.
What Playwright MCP actually does
Open a project, go to Test Tools, and pick Playwright MCP. Choose the test cases you want executed (or write a freeform instruction in plain English), point it at your app's URL, and hit Start Session. A real browser opens on Arqo's infrastructure and an AI agent begins working through your cases.
- It drives a genuine Chromium browser — the same engine behind Playwright — not a simulation
- Every action streams to a live view, updating roughly once a second, so you can watch it work
- Results (Pass / Fail / Blocked) and notes are written straight into an Arqo test run
- It captures screenshots as evidence, automatically grabbing one whenever a case fails
- You can pause, resume, or open the browser full-screen at any point
It hands control back for logins and CAPTCHAs
The hardest part of AI browser automation isn't clicking buttons — it's the walls: login forms, two-factor prompts, 'prove you're human' checks. Playwright MCP doesn't try to fake its way past them. When it hits one, it stops, shows a take-over banner, and lets you interact with the live browser directly. Your clicks and keystrokes are forwarded to the real page. Once you're through, you hand control back and the agent resumes exactly where it left off.
This human-in-the-loop model is what makes it trustworthy. The agent never guesses a password, never claims a case passed when it was actually blocked, and never fabricates a result to get unstuck.
Note: Sign in once with 'Save login after I sign in' and Arqo securely stores the browser session (encrypted at rest, never shown or sent to the AI) so future sessions start already authenticated.
Freeform mode: describe a flow, watch it run
Beyond executing existing cases, Playwright MCP has a freeform mode. Type something like 'log in as a standard user, add a product to the basket, and complete checkout as a guest' and the agent carries it out end to end. It's the fastest way to smoke-test a new flow before you've written formal test cases — and a great way to turn a working flow into cases afterwards.
Run it straight from a test run
Every test run in Arqo has a 'Run with Playwright MCP' button. Click it to execute that run's cases in a real browser, with the AI writing pass/fail results and screenshots back into the very same run. Manual and AI execution live side by side in one place — no separate dashboard, no CSV shuffling between your CI pipeline and your test management tool.
Why this matters for modern QA
AI-assisted testing has mostly meant generating test cases. Executing them has stayed manual or stayed locked behind code. Playwright MCP is the missing half: autonomous, agentic execution of real test cases in a real browser, with a human on hand for the moments that genuinely need one.
- Coverage without maintenance — no scripts to break when the UI changes
- Every QA engineer can automate, not just those who write Playwright
- Evidence by default — screenshots and notes attached to each result
- Results flow into the same reports and PDF exports as your manual runs
Best practice: point it at staging
For the cleanest runs, aim Playwright MCP at your own application's staging environment with CAPTCHA disabled — not at live third-party sites like Google, whose anti-bot systems will challenge any datacenter browser. On your own app, the agent runs uninterrupted end to end.
Getting started
Playwright MCP is available on Plus and Team plans and is currently in Beta, with monthly session limits per plan. Open any project, head to Test Tools, and select Playwright MCP to run your first AI-driven browser session. The test cases you've already written are all you need.