开发者 GitHub 用户 Tthanh2k6 发布了名为 turbo-browser-mcp 的基于 Playwright 的 Model Context Protocol (MCP) 服务器,旨在通过批量处理工具减少大语言模型(LLM)在浏览器自动化任务中的 token 浪费与多轮交互延迟 1。该工具将原本需要调用的 25 至 33 个细粒度浏览器原语整合为 10 个批量工具,并支持 DOM 增量快照、单轮表单预填以及直接从磁盘上传文件等核心功能 1。
基准测试数据显示,通过消除 10 次往返交互,在处理包含 10 个字段的表单时,浏览器执行时间从 574 毫秒显著降低至 82 毫秒,同时节省了约 71,800 个用于重放历史记录的 token 1。为保障安全,turbo-browser-mcp 默认启用“Human-in-the-loop Submit guard”机制,除非获得明确允许,否则阻止自动提交表单 1。此外,该项目针对 Chrome 136+ 和 144+ 版本提供了 CDP Keep-Alive Bridge Daemon,以避免远程调试弹窗造成的阻塞,并支持 auto、cdp、persistent 和 launch 四种连接模式 1。目前,该工具已通过分布在 86 个套件中的 407 个测试用例验证 1。
A developer identified as Tthanh2k6 on GitHub has released turbo-browser-mcp, a Model Context Protocol (MCP) server built on Playwright that aims to reduce token waste and latency in large language model (LLM) browser automation 1. The tool consolidates between 25 and 33 granular browser primitives into just 10 batch tools, enabling features such as delta DOM snapshots, single-round form pre-filling, and direct disk file uploads 1. By integrating these functions, the project seeks to address the inefficiencies inherent in multi-step interactions where each small action traditionally requires separate processing cycles 1.
Performance benchmarks indicate significant improvements when handling complex tasks, such as filling out forms with ten fields 1. The system reduces browser execution time from 574 milliseconds to 82 milliseconds by eliminating ten round trips, while also saving approximately 71,800 tokens associated with replaying historical context 1. To ensure safety during automated operations, the server includes a default Human-in-the-loop Submit guard that prevents automatic form submissions unless explicitly permitted 1. Additionally, the tool offers four connection modes—auto, cdp, persistent, and launch—and provides a CDP Keep-Alive Bridge Daemon specifically for Chrome versions 136+ and 144+ to avoid blocking from remote debugging pop-ups 1. The project’s robustness is supported by 407 passing test cases distributed across 86 suites 1.
评论
还没有评论,欢迎留下第一条。