MCP 浏览器使用

通过此MCP可以实现什么

该项目旨在利用模型上下文协议（MCP）和Selenium，使AI代理能够执行网页使用、浏览器自动化、抓取和自动化。

我们的使命是让AI代理在最少的人类监督下完成任何网络任务——所有这些都是基于自然语言指令。

如何使用此MCP

请参考modelcontextprotocol.io上的MCP文档。

请注意，您需要在您的MCP配置文件所指向的Python环境中安装所有依赖项。例如，如果您指向python或python3可执行文件，那么您将指向全局Python环境。通常建议指向一个虚拟环境，如：

/Users/yourname/code/mcp_browser_use/.venv/bin/python

如果您已将此仓库克隆到本地的code文件夹中，您的MCP配置文件应如下所示：

{
    "mcpServers": {
        "mcp_browser_use": {
            "command": "/Users/janspoerer/code/mcp_browser_use/.venv/bin/python",
            "args": [
                "/Users/janspoerer/code/mcp_browser_use/mcp_browser_use"
            ]
        }
    }
}

它将位于这里（在macOS中）：/Users/janspoerer/Library/Application Support/Claude/claude_desktop_config.json。

请参阅requirements.txt以查看需要安装哪些依赖项。

重启Claude以检查JSON配置是否有效。如果有问题，Claude会引导您查看MCP的错误日志。

如果设置成功，在Claude的“新聊天”窗口右下角将会看到一个小锤子图标。锤子旁边将显示MCP提供的功能数量。

点击锤子，您将看到类似以下内容：

Available MCP Tools

Claude can use tools provided by specialized servers using Model Context Protocol. Learn more about MCP.

click_element
Click an element on the page. Args: session_id: Session ID of the browser selector: CSS selector, XPath, or ID of the element to click selector_type: Type of selector (css, xpath, id)

From server: mcp_browser_use

close_browser
Close a browser session. Args: session_id: Session ID of the browser to close

From server: mcp_browser_use

fill_text
Input text into an element. Args: session_id: Session ID of the browser selector: CSS selector, XPath, or ID of the input field text: Text to enter into the field selector_type: Type of selector (css, xpath, id) clear_first: Whether to clear the field before entering text

From server: mcp_browser_use

navigate
Navigate to a URL. Args: session_id: Session ID of the browser url: URL to navigate to

From server: mcp_browser_use

scroll
Scroll the page. Args: session_id: Session ID of the browser x: Horizontal scroll amount in pixels y: Vertical scroll amount in pixels

From server: mcp_browser_use

send_keys
Send keyboard keys to the browser. Args: session_id: Session ID of the browser key: Key to send (e.g., ENTER, TAB, etc.) selector: CSS selector, XPath, or ID of the element to send keys to (optional) selector_type: Type of selector (css, xpath, id)

From server: mcp_browser_use

start_browser
Start a new browser session. Args: headless: Whether to run the browser in headless mode

From server: mcp_browser_use

take_screenshot
Take a screenshot of the current page. Args: session_id: Session ID of the browser

From server: mcp_browser_use

演示视频 (YouTube)