Playwright MCP开源MCP服务器，利用_playwright让LLM控制真实的浏览器以及可访问性快照

4.8 (6)

审阅者 Daniel Nikulshyn·更新 2026年5月

Open Source Browser Automation MCP Developer Tools LLM Agents Web Scraping Testing

概览

Playwright MCP 是一个开源的 Model Context Protocol 服务器，它将 Playwright 的浏览器自动化能力暴露给大型语言模型。它不依赖截图和视觉模型，而是提供网页的结构化无障碍快照，为代理提供快速、确定性的 DOM 视图，代理可以在此基础上进行推理和操作。它让基于 LLM 的代理能够在 Chromium、Firefox 和 WebKit 上浏览网站、点击元素、填写表单、提取数据并执行端到端工作流程。由于它支持 MCP，它可以与任何兼容的客户端（如 Claude Desktop、Cursor 或自定义代理框架）无缝集成，使真实浏览器任务能够被自主或协助工作流程所使用。

主要功能

MCP服务器接口适用于LLM代理
结构化的可访问性树快照
_playwright跨浏览器支持
点击，类型，导航和表单填写动作
无头或全页面浏览器模式
与Claude，Cursor和自定义客户端集成

价格

模型: Free
分类: MCP Servers
评分: 4.8 / 5 (6)

使用场景

自治Web导航为AI代理

在Claude Desktop，Cursor或自定义框架内启用LLM代理，可以使用确定性的可访问性树快照浏览网站，点击链接，完成多步任务而不需要截图

自动化表单填写和工作流

让代理在 Chromium，Firefox和 WebKit中登录门户，填写表格，执行端到端工作流程，而无需编写自定义自动化脚本

结构化Web数据提取

使用可访问性树快照可靠地从网页内容提取并给予LLM快速token有效的DOM.view用于剿采和研讨任务

端到端测试与LLM协助

将_playwright的跨浏览器自动化与LLM结合进行交互式MCP兼容客户端进行生成，运行和调试端到端测试

优点 & 缺点

优点

免费和开源
可靠的自动化不需要截图
适用于任何兼容MCP的客户端
_playwright通过多个浏览器引擎支持
快速，确定性的可访问性基于交互
需要技术设置和Node环境

缺点

需要技术设置和 Node 环境
MCP 生态系统仍在发展中
在复杂页面上，令牌使用量可能增加
仅适用于纯视觉任务的效用有限

评测

4.8

6 个评分的平均值。

登录以留下评测。

Devin Walker

Mar 1, 2026

Does the job

Pretty happy overall. Structured accessibility tree snapshots just works and works with any MCP-compatible client. but no dealbreakers — I'd recommend it to a friend without hesitating.

Tariq Aziz

Feb 17, 2026

Solid for our team

We rolled this out across the team last quarter and free and open source. MCP server interface for LLM agents fits neatly into how we already work, and integration with Claude, Cursor, and custom clients removed a step we used to do by hand. but it has held up under daily use.

Ahmed Saleh

Dec 30, 2025

Does the job

Pretty happy overall. Integration with Claude, Cursor, and custom clients just works and fast, deterministic accessibility-based interactions. MCP ecosystem still maturing can be annoying, but no dealbreakers — I'd recommend it to a friend without hesitating.

Rina Desai

Sep 21, 2025

Use it every day

Honestly didn't expect to like it this much. MCP server interface for LLM agents is exactly what I needed, and supports multiple browser engines via Playwright. but I reach for it almost every day now and it just clicks.

Sofia Lindqvist

Aug 15, 2025

Compared a few options

Evaluated this against two competitors. Where it wins: headless or headed browser modes and free and open source. Where it lags: mCP ecosystem still maturing. On balance the feature set — especially click, type, navigate, and form-filling actions — justifies the 4 stars for our use case.

Linda Petersen

Jun 12, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on structured accessibility tree snapshots, and works with any MCP-compatible client caught me off guard. Limited usefulness for purely visual tasks is why this isn't a perfect score, still, I'd recommend giving it a real trial.

问答

Is it suitable for visual or screenshot-based testing tasks?

Not really. Playwright MCP is designed around structured accessibility tree snapshots rather than screenshots, which makes it fast and deterministic for DOM-based actions but limits its usefulness for purely visual tasks like pixel comparisons.

How much does Playwright MCP cost and what's the licensing?

Playwright MCP is free and open source, so there are no license fees. You only pay for the underlying compute to run the Node environment and any LLM tokens consumed by your MCP client when interacting with pages.

Which LLM clients and browsers does it work with?

It works with any MCP-compatible client, including Claude Desktop, Cursor, and custom agent frameworks. For browsers, it supports Chromium, Firefox, and WebKit via Playwright, in either headless or headed modes.