TestFinch

BlogFlowQA

Playwright MCP for testing: when your coding agent's browser is enough, and when you want kept tests

Playwright MCP for testing is free, open source, and enough when your coding agent only needs to look at the change it just made. FlowQA is for the tests you keep and run after every deploy. Most teams building with agents can use both.

Use Playwright MCP for testing when your coding agent needs to open the page it just changed, click through it once, and tell you what it saw. Use FlowQA when you want that same journey kept as a test that runs after every deploy, with nobody at the keyboard and no agent session open. Playwright MCP costs nothing, and you can use both.

What Playwright MCP does well#

As of 11 November 2026, the Playwright MCP repository describes it as "a Model Context Protocol (MCP) server that provides browser automation capabilities using Playwright". It is Microsoft's project, published under the Apache-2.0 license, so it is free and open source. The same README says:

  • The agent reads pages as accessibility snapshots, not screenshots, so it needs no vision model.
  • Install guides cover Claude Code, Cursor, Codex, VS Code and many other clients, and it needs Node.js 18 or newer.
  • The browser keeps a persistent profile by default, so a sign-in survives between sessions; --isolated starts clean each time.
  • Opt-in tools generate a locator for an element and verify an element, a list, a text or a value.
  • Recorded actions can come back as Playwright code, in TypeScript by default.

Your agent gets a real browser on your own computer, it works against localhost, and there is nothing to sign up for.

The README also says a coding agent "might benefit from using the CLI+SKILLS instead", because CLI workflows avoid loading large tool schemas and accessibility trees into the model's context. It says MCP stays relevant for exploratory automation and long-running autonomous workflows.

Where Playwright MCP for testing stops#

Playwright MCP gives an agent a browser for the length of a session.

Nothing is kept unless the agent writes it down. A walk through checkout tells you checkout works now. Tomorrow's change gets no check unless someone asks again, or the agent wrote a Playwright test file and your team runs and maintains it.

Each walk is a new decision. The agent chooses its clicks each time from what the page shows. A regression test needs the same steps and the same checks on every run, so a failure means the product changed.

Someone has to start it. A browser driven from a coding session runs when the session does, not on a schedule or on every deploy.

If your agent writes Playwright test files, a developer owns them from then on; Playwright vs record-and-replay covers where that line falls.

What FlowQA adds#

FlowQA keeps tests and runs them. Its AI agent maps your site and proposes a plan of journeys for you to review, or your coding agent writes the tests through the TestFinch MCP connector. A test from an exploration is kept only once it has passed twice on fresh sessions and then failed on a page broken the way its assertions describe. That shows the test can fail, not that it checks the right thing.

Tests then run on a schedule, from your pipeline, or from a GitHub deploy hook that reports a check on the commit. When a test fails on a changed page, FlowQA proposes one small repair, replays it twice on fresh sessions, and waits for your approval.

The connector for coding agents works with Claude Code, Codex and Cursor. Your agent can save up to 50 tests in one call, run them on a cloud browser, read the failing step and its screenshot, and review proposed repairs. The TestFinch CLI runs the same tests against localhost.

The connector does not hand your agent a live page to click around in. Its tools save, run and read tests, so for looking at a page while building, Playwright MCP is the better tool.

FlowQA is also web only and runs tests in Chromium, with emulated phone and tablet sizes. Its tests are records in your workspace, not code in your repository.

Side by side#

Playwright MCP's column is from its repository, checked on 11 November 2026.

Playwright MCP FlowQA
What you get A browser your coding agent drives Kept tests, run for you
Price Free, open source (Apache-2.0); you pay for the model your agent uses Free plan: 1 project, up to 100 tests, 1,000 cloud test runs a month; Starter is $49 a month
Runs on Your computer Cloud browsers, or your computer for free
What stays afterward Whatever the agent wrote down The test, its runs and their screenshots
Runs with no agent session No Yes: schedules, CI, GitHub deploy hooks
When the page changes The agent adapts on the next walk One repair is proposed and waits for approval

Using both#

A workable loop for a team that builds with Claude Code or Cursor:

  1. While building, the agent opens the change through Playwright MCP and fixes what it sees.
  2. Once the journey works, the agent saves it as a FlowQA test and runs it once on a cloud browser.
  3. From then on the test runs on every deploy, and a failure names the step and shows the page.

Pick Playwright MCP if#

You want your agent to check its own work before you commit, you are exploring or reproducing something once, or you want the agent to write Playwright code that a developer will own.

Pick FlowQA if#

You want the journeys that matter kept as tests that can fail, run after every change without a prompt, and repaired only with your approval.

FlowQA explores your site, writes the tests, keeps each one only once it fails on a page broken the way its checks describe, and runs them after every change.

See FlowQA