Playwright MCP Server

What it does

Playwright MCP gives a model browser automation through Playwright. It works from structured accessibility snapshots, so a vision model is not required, and tool actions target elements deterministically from the page structure. The README contrasts MCP for persistent, stateful agentic loops with the Playwright CLI for token efficient coding agent workflows.

The standard install runs with npx @playwright/mcp@latest and needs Node.js 18 or newer. The server can run headed or headless, pick chrome, firefox, webkit or msedge, emulate devices, and connect to an existing browser through CDP or the Playwright extension.

Browser state can persist in a user data directory, run isolated in memory, or start from a storage state file. A JSON config file can set browser, server, capabilities and context options.

Configuration

Run with npx @playwright/mcp@latest. Node.js 18 or newer is required, and a browser can be installed through Playwright if one is missing.

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Source: https://github.com/microsoft/playwright-mcp. Placeholders only. Replace them in your own environment, and never commit real keys to a repository.

Verified tools and features

• browser_navigate Config checked

  Navigates the browser to a URL.

• browser_snapshot Config checked

  Returns the page accessibility tree with element references for later actions.

• browser_click and browser_type Config checked

  Clicks an element and types text using accessibility references.

• browser_fill_form Config checked

  Fills multiple form fields in one call.

• browser_take_screenshot Config checked

  Captures a screenshot of the current page.

• browser_evaluate Config checked

  Executes JavaScript in the page context.

Security notes

Stacks featuring this server