💡

Microsoft's official Playwright MCP server. Drives the browser via the accessibility tree for fast, deterministic automation without vision models.

⚙️
Browser test automation Automation
Installation · plugin marketplace (claude-plugins-official)
$ /plugin install playwright

Microsoft official MCP server

What it does

Playwright MCP gives LLMs a structured way to drive a real browser. Instead of screenshotting pages and asking a vision model to click, it exposes the accessibility tree — so the agent sees semantic roles and names (button, textbox, link) and can act deterministically.

Features

  • Accessibility-tree interactions — locate and act on elements via DOM structure, not pixels
  • No vision model required — uses structured data only; fast and cheap
  • Deterministic tools — removes screenshot ambiguity for consistent outcomes
  • Browser automation — navigation, clicks, form fills, file uploads, keystrokes
  • Code generation — convert the actions you performed into Playwright test code

Use cases

  1. Describe a user flow in natural language and Playwright MCP runs it as an E2E test.
  2. Automate form fills, button clicks, and page navigation to accelerate manual QA.
  3. Export the browser actions as Playwright test code and wire into CI/CD.

When to use it

  • Author deterministic E2E tests in plain language
  • Automate flaky manual QA steps
  • Scrape structured content from pages where the accessibility tree matches what humans see

Frequently Asked Questions

What is Playwright MCP?

Microsoft's official Playwright MCP server. It drives the browser via the accessibility tree for fast, deterministic automation — no screenshots or vision models needed.

Where does it run?

Claude Code, Claude Desktop, VS Code, and Cursor.

How do I install it?

In Claude Code, run `/plugin install playwright`. It's listed on the claude-plugins-official marketplace.

How is it different from screenshot-based tools?

It targets the DOM accessibility tree rather than pixels, so it's fast and deterministic without vision models. That eliminates the ambiguity of screenshot-based selection and produces consistent results.

What can it do?

Accessibility-tree-based element interaction, browser automation (navigation, clicks, form input, file uploads), and automatic conversion of performed actions into Playwright test code.