OpenCode 推出了名为“OpenCode Browse”的新工具,该工具包含一个 MCP 服务器和一个 Chromium 浏览器扩展 1。其核心功能是允许 AI 代理使用用户的真实浏览器环境,利用现有的登录状态、标签页和扩展进行网页浏览、点击、输入及截图等操作 1。
在操作流程中,代理会在后台打开分组标签页读取页面并执行操作;当遇到需要用户交互的情况(如登录、双重验证或支付)时,系统会将控制权交还给用户,待用户完成操作并点击“Continue”后,代理继续执行任务 1。该工具支持 Chrome、Edge、Brave、Arc、Helium、Vivaldi、Opera 及 Chromium 等多种浏览器 1。
在集成方面,OpenCode Browse 支持 OpenCode、Claude Code、Cursor、Codex、Zed 等任何 MCP 客户端 1。安装方式包括首先安装 Chrome 扩展,或通过 curl、pwsh、npm、bun 安装 CLI 1。
OpenCode has released a new tool called "OpenCode Browse," which consists of an MCP server and a Chromium browser extension. 1 This utility allows AI agents to operate within the user's actual browser environment, leveraging existing login sessions, open tabs, and installed extensions for tasks such as web navigation, clicking, and taking screenshots. 1
The tool functions by opening agent-controlled tabs in the background, where it can read pages, perform clicks, enter text, and capture images. 1 When interactions requiring human intervention arise, such as logging in, two-factor authentication, or payments, the system hands control back to the user; once the user clicks "Continue," the agent resumes its task. 1
"OpenCode Browse" supports a wide range of browsers including Chrome, Edge, Brave, Arc, Helium, Vivaldi, Opera, and Chromium. 1 It is compatible with various MCP clients such as OpenCode, Claude Code, Cursor, Codex, and Zed. 1 Users can install the tool by first adding the Chrome extension or by installing the CLI via curl, pwsh, npm, or bun. 1
评论
还没有评论,欢迎留下第一条。