结果：

Task Completed

The Playwright MCP server installation is currently in progress via npm install. The server is already properly configured in cline_mcp_settings.json with server name "github.com/executeautomation/mcp-playwright". Once the installation completes, we can demonstrate the server's capabilities by using one of its tools.
 

(base) PS D:\A\1Junior\Cline> history                                                                         

  Id CommandLine
  -- -----------
   1 try { . "c:\Users\Eric1\AppData\Local\Programs\Microsoft VS Code\resources\app\out\vs\workbench\contr... 
   2 npm install -g @executeautomation/playwright-mcp-server
   3 npm install -g @executeautomation/playwright-mcp-server
   4 npm install -g @executeautomation/playwright-mcp-server
   5 npx -y @executeautomation/playwright-mcp-server
   6 npm config set PLAYWRIGHT_DOWNLOAD_HOST=https://npmmirror.com/mirrors/playwright
   7 $env:PLAYWRIGHT_DOWNLOAD_HOST="https://npmmirror.com/mirrors/playwright"
   8 npm install -g @executeautomation/playwright-mcp-server
   9 npm list -g @executeautomation/playwright-mcp-server
  10 npx -y @executeautomation/playwright-mcp-server
  11 #安装playwright
  12 pip3 install playwright
  13 python -m playwright install
  14 npm install -g @executeautomation/playwright-mcp-server

然后换用R1类深度思考模型，我只能说过于耗时且依旧会绕进去。 

搜搜搜。搜到这个，

Windows环境下搭建Cline+Playwright+MCP，超详细_playwright mcp环境搭建详细步骤-CSDN博客

我感觉这个有可能是“失败”的原因，它变成了一个端口，所以你会一直等而无所得。

 

 这里图中的config为：

{
  "mcpServers": {
    "playwright": {
      "url": "http://localhost:8931/"
    } 
  }
}

注意，千万不要让AI获取，这里幻觉异常严重。

🆗了

 

长思考模型做这种长上下文时效性真的是烂透了。 

 

换用V3，截图成功，但是依旧404

查找原因，因为并没有成功调用MCP server的，而是VScode自带的。

这次使用V3,并明确指出使用playwright mcp server. 但是成功进入了搜索框输入步骤，但是缺乏后续。再一次卡住。卡住后，它又回去使用browser，然后就只能截图baidu主页就404.

这样是正常的。

 很好，改用R1模型虽然慢，但是一步一步到搜索MCP这一步了，成功调用了click组件。

 红温了，暂告一段落。

使用MCP playwright, 打开百度，搜索MCP, 获取目标网页的截图。 

 其它模型，依旧是指定了，也不会用

后续：使用中转 claud-3.5-sonnet 成功实现全步骤，但是很贵，这一个task就用掉我4块钱的token. 显然及其昂贵且不合理。