此網站正在建設中。

Playwright MCP

MCP server from Microsoft that lets AI models control a browser with Playwright

Playwright MCP is a Model Context Protocol (MCP) server from Microsoft. It lets AI models control a web browser through Playwright. The model reads a structured accessibility snapshot of the page, where each element has a reference, and uses it to click, type, fill forms and navigate. No vision model or screenshots are needed.

It runs locally with Node.js 18 or newer (npx @playwright/mcp@latest) and works with MCP clients such as VS Code, Cursor, Windsurf, Claude Code and Claude Desktop. It supports Chrome, Firefox, WebKit and Edge. It can also serve over HTTP with the --port option, run from a Docker image (headless Chromium only), or connect to your open Chrome or Edge tabs through the Playwright Extension.

The docs list more than 70 tools. Extra groups such as network mocking, storage, DevTools, PDF, coordinate-based input and test assertions are turned on with --caps.

Free and open source under the Apache 2.0 licence.

適用於

Developers who connect AI assistants and agents in MCP clients such as VS Code, Cursor or Claude to a real browser. The maker suggests it for agent loops that keep browser state, such as exploratory automation and self-healing tests.

使用案例

Let an AI agent open websites, click and fill in forms
Generate Playwright tests by letting AI explore a site
Run exploratory checks and self-healing tests
Mock network requests and read console messages
Save pages as PDF or take screenshots
Let AI work in your logged-in Chrome tabs via the extension

替代方案：

Chrome DevTools for agents
Supports Chrome, Firefox, WebKit and Edge and can run from a Docker image; Chrome DevTools for agents works with Chrome only.
Firecrawl MCP
Free and runs locally with no API key or credits; Firecrawl MCP uses paid Firecrawl API credits beyond the free 1,000 a month.

Playwright MCP的替代選項 3

MCP 伺服器

Chrome DevTools for agents

MCP server and CLI that lets coding agents control and inspect a live Chrome

開源 API / CLI

Records performance traces and runs Lighthouse audits with source-mapped console messages from Chrome DevTools; it also ships a CLI for use without MCP.