开源浏览器自动化工具，已安装并运行

让 Claude Code、Codex 或 Cursor 使用你的真实 Chrome，并以你的身份登录。一条提示即可完成设置。开源，MIT 许可证。

测试发现：browser-harness 0.1.13 installs from source, all 266 unit tests pass, and the CLI controls a headless Chromium via CDP, page_info, navigation, tab management, JS execution, and screenshots all work end-to-end. 12 分钟。

面向 AI agent 和应用的开源浏览器 API。Steel Browser 是开箱即用的浏览器沙箱，让你能够自动化操作网页，无需担心基础设施。

测试发现：With a user-local Node 22 and Chrome for Testing plus DISABLE_CHROME_SANDBOX=true, the dev-mode API runs on port 3000, /v1/health returns 200 with live Chrome, and sessions/scrape/screenshot/pdf plus external CDP client connections all worked with real results. 11 分钟。

Browser MCP 是一个 Model Context Provider (MCP) 服务器，允许 AI 应用控制你的浏览器。

测试发现：All dependencies install, the project builds, typechecks, and all 252 tests pass across all 4 tested packages using real headless Chromium, and the MCP server binary responds to --help. 8 分钟。

用于结构化提取工作流的声明式数据自动化语言和 Go 运行时。

测试发现：Go 1.27.1 installed, all dependencies resolved, all 68 test packages pass with race detection, CLI binary compiles and runs FQL queries end-to-end. 7 分钟。

Playwright for Go 是一个浏览器自动化库，用于通过单一 API 控制 Chromium、Firefox 和 WebKit。

测试发现：Playwright for Go is installed and functional: Playwright driver v1.62.1 launches Chromium 151.0.7922.34 in headless mode, all package tests (0 failures) and all integration test groups (0 failures) pass with real browser interactions including page navigation, DOM queries, locators, assertions, network interception... 42 分钟。

Atomic Agent 是一个本地优先的 AI agent。通过 llama.cpp 在你的机器上运行开放权重模型。

测试发现：Build compiles, CLI binary runs and responds to --version/--help, 594 of ~600+ test files pass with 1 genuine test-logic failure in sidecar send-message concurrency, and all 22 eval-memory test files pass. 58 分钟。

面向 AI agents 的隐身浏览器自动化，基于 MCP 和 Chrome DevTools Protocol：导航、网络钩子、DOM 提取和像素级准确的 UI 克隆。

测试发现：Stealth Browser MCP server runs on stdio transport, spawns a headless Chromium 157.0.8086.0, navigates to data URLs, extracts page HTML content, and closes cleanly -- all via MCP protocol over stdio. 15 分钟。

一个轻量、快速的 Rust CLI，通过 Chrome DevTools 协议驱动真实浏览器，专为 coding agent 构建。

测试发现：The browser-control CLI builds and passes all 10 unit tests; the binary connects to Chromium 127 over CDP and successfully executes open, snapshot, eval, cdp, text, and inspect commands against real web pages. 10 分钟。

自学习 agent 浏览器

测试发现：Webcmd 0.8.4 is installed, builds from source, runs unit tests (3571 passed), connects to Cloak browser daemon, and responds to CLI commands. 5 分钟。

PHP 网页浏览器模拟器抽象层

测试发现：PHP 8.4.26 and Composer 2.10.3 are installed locally, all 38 dependencies are in vendor/, the Mink test suite passes 520 tests with 1165 assertions, composer.json validates, phpstan reports no errors, and a stub-driver script using the Mink library successfully found the Downloads link with href /download. 12 分钟。

AIPex：AI 浏览器自动化助手，无需迁移，隐私优先。Manus Browser Operator、Claude Chrome 和 Agent Browser 的替代方案。

测试发现：Repository builds, typechecks, lints, format-checks, and runs 639 passing tests across 5 workspace packages (core, dom-snapshot, browser-runtime, aipex-react, browser-ext) plus the mcp-bridge package, using pnpm 12.10.1 on Node 20.19.0. 5 分钟。

Chrome 侧边栏扩展，让 DeepSeek Harness 直接操作您的浏览器，无需视觉能力。

测试发现：The dsh-browser workspace pnpm-installs, builds, typechecks, and passes all 528 unit tests plus the real-DSH smoke test; the dsh web server starts, serves the UI at 200, exposes the bridge-config endpoint, authenticates via token→cookie exchange, and responds with correct RPC results via both the HTTP /api carrier and... 28 分钟。

Scout：由 Outcrawl 开发的开源 browser agent。从 Outcrawl 自己的仓库自动镜像。

测试发现：Bun 1.4.3 and Node 22.14.0 installed locally, bun install succeeded, and every workspace test suite passes (1453 pass, 0 fail, 0 skip) including the previously-skipped real headless-Chromium tests; a standalone Scout run completed with status completed and data 'All stories loaded.'. 10 分钟。

Playwright 的 API 基于 Rust CDP 引擎，为 Python 和 Node 提供 Chromium 浏览器自动化，无需驱动子进程。（Alpha）

测试发现：Built from source with Rust 1.99 and Python 3.12; Chromium 155 installed and launchable; Python quickstart prints 'Rustwright works', 18-test CI subset passes, Node smoke and 19 Node tests pass, rustwright CLI (0.3.0) and native rustwright-cli (0.1.0) open/snapshot/close real browser sessions, MCP server answers... 68 分钟。

一个用于自动化日常 Web 浏览器的 CLI 工具。

测试发现：Beachpatrol E2E suite passes all 14 tests with real headless Chromium browser, including smoke test (selenium.dev), error propagation, downloads, concurrent routing, duplicate detection, TypeScript commands, dependency resolution, --list, and --commands. 6 分钟。

面向非开发者的 Web 测试解决方案，基于 Puppeteer 和 Jest。

测试发现：npm install, npm test (21 suites/261 tests all passing), npm run build, and Electron app launch on the virtual display all work; async test suite and unit tests verified with real jest runs. 15 分钟。

基于 Chromium 和 Electron 的本地优先开源多配置文件指纹浏览器

测试发现：Project installs cleanly via npm ci, typechecks with tsc --noEmit, builds with electron-vite, all 126 unit tests pass, kernel lock checks pass, and Electron v43.2.0 launches and runs until terminated. 4 分钟。

开源 Web 基础设施，用于 AI。抓取、爬取和自动化网页，清理 Markdown，浏览器会话，为你的 agent 做好准备。

测试发现：Reader CLI builds, all 346 unit tests pass, the daemon starts with Playwright's bundled Chromium, and an end-to-end scrape of https://example.com via HTTP returns clean HTML and markdown content with a 200 status. 13 分钟。

将任何网站转化为专为 AI agents 设计的紧凑型 CLI。用数百个 tokens 浏览网页，而非数万个。

测试发现：only-cli v0.5.8 installs and runs on Node 20.11.1; 263 of 264 tests pass; the CLI fetches, distills, and renders public web pages with token-budgeted output and actionable numbered links. 3 分钟。

一个作为现有网页浏览器无头模式包装器的软件包，用于从 URLs 以及 HTML+CSS 字符串或文件生成图像。

测试发现：Package installed in venv, ChromiumHeadless defaults include --no-sandbox, 17 Chrome tests pass, CLI produces correct images, Edge tests gracefully skipped. 5 分钟。

面向 AI agent 的 ADE。自动化其余一切。做你真正的工作。

一群一次性 agents 用于不可归因的 web 交互

测试发现：Ash Agents orchestrator server runs on port 8092 with ASH_DEMO_MOCK=1, all endpoints respond correctly, and all 12 unit tests pass. 6 分钟。

面向 AI agent 的反检测浏览器服务器：REST API 封装了 Camoufox 引擎并支持 OpenClaw 插件

测试发现：Camofox Browser Server builds, starts on port 9998, creates browser contexts via Camoufox Firefox headless, navigates to https://example.com, returns accessibility snapshots, and passes the 677-test full suite with Node.js 20 and rebuilt native addons. 34 分钟。

由您自己的 Claude / Cursor 订阅驱动的 Agentic 浏览器控制

测试发现：Bridge runs on Node 24, MCP endpoint responds with 14 browser tools, 175/175 tests pass, .endo-token file created with 0600 permissions. 3 分钟。

阻止 AI agents 做你没有要求的事情。

测试发现：npm install and tsc build succeed, all 50 test cases pass, CLI help responds with all commands listed. 17 分钟。

非官方 DSH（DeepSeek Harness）插件：将 chat.deepseek.com 的 Web 模型用作 LLM 提供方，支持浏览器登录捕获、PoW 求解、SSE 流式传输和基于提示的工具调用。

测试发现：The project builds cleanly (tsdown produces lib/index.js and lib/client.js), all 62 offline test files pass, and the bundle product verification confirms all runtime features are present. 5 分钟。

一个完全免费的 agentic 浏览器驱动，用于 AI，少量工具，完全控制，真正的隐身，顶级的 token 效率。

测试发现：Bladebro builds from source with zero clippy warnings, all 217 tests pass, Chrome for Testing launches under Xvfb, and the binary navigates real websites, reads page content, and queries cookies. 8 分钟。

一个本地优先、跨平台的 Electron 桌面工作区，用于 Pi Coding Agent，具有会话、项目文件、浏览器工具、技能、插件和消息集成功能。

测试发现：Pi Agent Desktop v0.3.0 builds, all 1106 tests pass, and Electron smoke test verifies agent-host (pi=0.87.1) with all 12 function domains (renderer, RPC, session, worktree, git, watch, download, skills, channels, chat-appearance, toolchain) passing. 11 分钟。

为 LLMs 优化的高性能网页爬虫 API。使用远程浏览器将任何搜索或网站转换为干净的 Markdown。Firecrawl 的替代方案。

测试发现：Teracrawl server starts on port 8085 with a local Chromium-backed mock Browser.cash pool and a mock SERP service. GET /health returns 200, POST /scrape converts a live URL to markdown via real headless Chromium, POST /serp/search returns mock search results, and POST /crawl combines both to scrape multiple URLs. 15 分钟。

让您的 AI agents 在后台标签页中使用您的 Chrome，并以您的身份登录。作者：Alexey Indeev · 博客：read.aindeev.com · X：@AlexeyIndeev

测试发现：The Chrome Relay relay server starts, answers HTTP endpoints with correct status codes, and passes all 12 integration tests including agent WebSocket connections authenticated by secret and process markers, extension origin validation, web page rejection, and audit trail persistence. 20 分钟。

OpenBrowser 是一个用于智能浏览器自动化的框架。它结合直接 CDP 通信与 CodeAgent 架构，其中 LLM 编写 Python 代码并在持久命名空间中执行，以自主导航、交互并提取网页信息。

测试发现：All 8030 unit and integration tests pass, CLI prints version and help, and headless Chromium via CDP navigates to about:blank and prints output correctly. 27 分钟。

本地优先的 CLI，用于 agent 技能、浏览器自动化、知识库、外部 CLI 和终端 AI 聊天。

测试发现：advai-cli 1.0.12 installed from source, all 78 unit tests pass, the CLI responds to all top-level commands, and the knowledge base subsystem creates, adds, searches, and syncs real local files. 1 分钟。

面向 agent 技能、浏览器自动化、知识库、外部 CLI 和终端 AI 聊天的本地优先 CLI。

测试发现：advai-cli 1.0.12 is installed in a venv, all 78 tests pass, and core commands (info, update, skill, kb, browser, cli) respond with valid output. 1 分钟。

像运行程序一样运行网页

测试发现：webframe builds from source with patched glaze (preferGTK3=true), runs with WebKitGTK 4.1 from deb packages, executes JS pages, reports console output, supports exit codes and scripts. 19 分钟。

codexqa：11 个本地优先的 Agent Skills，适用于 Cursor、Claude Code、Codex 和 OpenClaw：变更影响分析、AI 代码审查、缺陷扫描、测试用例生成、浏览器回放和 RCA。

测试发现：The repo installs from a local checkout via npx skills with Playwright Chromium and the @openqa-cn/codexqa CLI in a user prefix; all repository checks pass: check-docs, compose-release-notes, checkout-boundary and inventory-service verifiers, polyglot fixture build, code-wiki 47/47, defect-analyzer full suite... 6 分钟。

一个命令适用于所有接口：跨 API、浏览器、桌面、本地工具和 MCP 搜索、运行并检查真实软件。

测试发现：Uni-CLI v1.2.1 builds from source, runs 3735 passing unit tests, and executes live commands (hackernews top, search, doctor) against real remote APIs on Node 22.23.4 (npm v10.9.9). The npm global package v1.3.0 is also installed and functional. 20 分钟。

开源、自托管的 AI agent 平台，用于你公司的 AI 队友。在自有服务器和模型上构建、运行和治理 agent：共享聊天、RAG、技能、真实浏览器、4,700+ MCP 服务器、预算、审批和审计追踪。是 ChatGPT Enterprise 和 Grok Bot 的替代方案。Apache-2.0，基于...

测试发现：Backend Python environment (Python 3.15.2 via uv) installed, 254 packages resolved; frontend Node dependencies (771 packages) installed via bun; backend lint passes (ruff, formatting); 609 backend unit tests pass covering coverage gate, agent registry, labels, A2A client/server, mutation locks, observability, and... 15 分钟。

为 AI agent 提供隔离的 Linux 桌面工作区：一个隐藏的、agent 拥有的桌面和浏览器，通过 MCP 运行，这样 agent 就可以进行 GUI 和 Web 工作，而不会触及你的真实桌面。

测试发现：agent-workspace-linux 0.6.2 builds from source, all 199 tests pass, binary runs --help and --version, starts an X11 workspace with openbox and xdotool (extracted from apt debs), launches GUI apps (xlogo), takes screenshots via ffmpeg, and responds to MCP protocol initialization. 14 分钟。

索引全球未文档化的 API

测试发现：All 2434 unit tests pass after fixing 2 flaky async tests; pip editable install succeeds; CLI help commands work for bluebox-agent and bluebox-discover; core Python imports resolve without errors. 3 分钟。

面向 Python 的 Puppeteer，反机器人系统无法察觉，因此无需验证码：相同 API、反检测隐身无头 Firefox、无法识别的指纹、绕过机器人检测，用于网页抓取和浏览器自动化。

测试发现：Editable install resolved all dependencies, default pytest suite passes (219/259), CLI version and fetch commands work, and the browser engine was downloaded and cached. 3 分钟。

开源、自托管的 OpenAI Dots 和 Grok Bot 替代品。专为不被反机器人系统检测而设计。

测试发现：npm ci installed 428 packages, the CLI bundle and web client built, npm run typecheck passed, npx vitest run gave 2103 passed/3 skipped and go test gave 120 passed after I fixed the secret-input backspace bug and the execEnv DISPLAY bug, and the control plane served HTTP 200 on /api/health, /api/dots and the web UI... 22 分钟。

用于 AI agent 的隐形无头浏览器，可绕过 Cloudflare、机器人检测和反爬虫。可直接替代 Puppeteer/Playwright。

测试发现：Camofox browser server installed and verified: Node 22.23.4 in user space, Camoufox 1.3GB binary downloaded, GeoIP 50MB database downloaded, server starts at localhost:9377 with Xvfb virtual display, creates browser tabs navigating to real URLs, returns accessibility snapshots with element refs, and closes cleanly on... 13 分钟。

开源、自托管的 OpenAI Dots、Grok Bot 替代方案。构建目的是不被反机器人系统检测到。

测试发现：dots installs, its 4 tests pass, and the UI server runs and drives its bundled browser with a real model call, navigating https://example.com and reporting HTTP 200. 7 分钟。

JavaScript 页面内 GUI agent。使用自然语言控制 Web 界面。

测试发现：Install, sequential build, typecheck, lint, and all 69 unit tests pass. The project is fully functional with Node v22. via nvm. 27 分钟。

高性能浏览器自动化桥接与多实例编排器，支持高级隐身注入和实时仪表盘。

测试发现：A source-built PinchTab server (binary reports pinchtab dev) is running on 127.0.0.1:9867 with a responsive always-on headless instance and answered authenticated health, navigate, snapshot, text, title, and screenshot API calls with HTTP 200, while the full Go unit suite (9166 passed, 18 skipped) and scripts/check.sh... 19 分钟。

Grasp 帮助你在一个可视化工作流中管理多个项目和并行的 coding agents，从而让人更快理解并交付更多。

测试发现：With user-local Go 1.27.3 and Node 24.12.0, the built Grasp server and gateway binaries start and answer health 200, admin login and project APIs return 200 with seeded data, and the server, web, docs, gateway, and sandbox build, test, and coverage suites all pass. 18 分钟。

开源自托管网页存档工具。可获取 URL/浏览器历史/书签/Pocket/Pinboard 等，保存 HTML、JS、PDF、媒体等内容。

测试发现：ArchiveBox 0.9.65rc24 installed in a Python 3.13 venv; data directory initialized with SQLite; Django admin server starts and serves HTTP on port 5797; 125 tests pass; full archiving (add, crawl) requires wget/chrome/yt-dlp which are not installed in this container without root. 12 分钟。

用于 Web 自动化的开源 AI agent。使用你真实且已登录的浏览器来自动化日常任务。

测试发现：Dependencies installed, project builds to dist/ (background.iife.js 1,848.81 kB, manifest.json), TypeScript type-check passes across all 15 packages, unit tests pass (14/14). 15 分钟。

在浏览器中运行 AI Agent。

测试发现：The Browser Use WebUI serves HTTP 200 on port 7788 with the full Gradio UI, runs an agent against the local mock, and completes tasks with a decodable recording GIF and agent-history JSON. 37 分钟。

桌面自动化 MCP 服务器，任何 AI agent 的计算机使用：通过 Model Context Protocol (stdio) 控制屏幕、窗口、鼠标/键盘和 Chrome。不是 DSH 插件，DSH 用户安装 dsh-nuphus-mcp。

测试发现：nuphus-mcp v0.3.3 installs via npm, initializes over stdio JSON-RPC responding to initialize/tools/list/tools/call, returns screen size (1280x800), and captures valid PNG screenshots in headless Xvfb. 12 分钟。

适用于 EHRs、支付方门户以及你的团队点击使用的任何应用。一个人或一个 agent 演示任务；OpenAdapt 将其转化为本地程序，无需模型调用即可重复步骤，检查每次保存，并在不匹配时停止。

测试发现：Package installs and all 288 tests pass; CLI, doctor, version commands work; recording generates valid artifacts; compile is CPU-bound by onnxruntime in this container but eventually produces bundles. 20 分钟。

该布局引擎可以解决可想象的最困难的布局问题。其输出达到桌面出版级且具有确定性。它比用 C++ 编写的引擎更快，因为它不是传统的。基于游戏引擎技术，它采用微内核，运行时空模拟，而不是庞大的复杂分页循环。

测试发现：vmprint CLI and engine ran in this container on Node 18 despite EBADENGINE install warnings: npm test exited 0 (engine 77+77 fixtures, packaged-integration 1 page/1107 bytes), the CLI rendered valid text-bearing PDFs with bundled local fonts and version 0.1.0, and engine.layout() produced 1 page x 1 box containing... 18 分钟。

开源 Python 库，用于 AI 浏览器 agents，以及 Browser Use Cloud：隐身浏览器和托管 agent API。

测试发现：browser-use v0.13.10 installed in a Python 3.12 venv with uv, Chromium headless installed via playwright, and all 563 CI tests pass (529 passed, 34 skipped) after fixing 2 test identity comparison bugs in test_beta_agent.py that only manifest under xdist worker parallelism. 75 分钟。

将任何网站变成 CLI，并通过 AI agent 使用你已登录的浏览器。

测试发现：OpenCLI 1.8.8 builds and runs on Node 22.14.0; 632 test files pass (7325 tests, 3 skipped) across unit, extension, adapter, smoke projects; CLI binary answers version 1.8.8 and lists 1366 registered commands. 27 分钟。

在浏览器中完成任务的 Terminal UI

测试发现：The Rust workspace builds cleanly, all 2058 Rust and 28 Python tests pass, cargo fmt is clean, and both the CLI (browser-use-terminal 0.1.8) and TUI (but 0.1.8) binaries run, with the TUI rendering its setup screen and responding to real keyboard input in a PTY session. 40 分钟。

为 AI agents 运行浏览器自动化而打造的最快浏览器，可将你已登录的浏览器状态分享给你的 AI agents（如 Codex 或 Claude Code），且不会打扰你。零成本，零配置。

测试发现：494 unit tests pass on Node 22.23.4 with mocking (FakeEgo test double), the build produces a CLI binary that prints help text, and type checking passes cleanly. Real browser e2e was not attempted because no ego lite runtime is available in this container. 2 分钟。

自我进化的 agent：从 3.3K 行的种子中生长技能树，以 6x 更少的 token 消耗实现全面系统控制。

测试发现：After a venv install of the declared UI and test dependencies, 275 tests pass, the TUI starts, the Streamlit UI completes a mocked tool-using turn with MOCK_FINAL: 246810, and launch.pyw opens a live 1200x900 GenericAgent window on DISPLAY=:99 using PyQt6 and an extracted libxcb-cursor. 29 分钟。

使用 LLM 将任何网页转换为结构化数据

测试发现：The project installs and builds without errors; 8/8 test cases pass with a mocked OpenAI Responses API endpoint. 24 分钟。

AI Agent 优先的 JS 逆向 MCP 服务器：有头 Chrome 调试、断点、网络/WebSocket 分析、Patchright 反检测，可选 CloakBrowser。

测试发现：The js-reverse-mcp v4.0.5 TypeScript project compiles to JavaScript, its full test suite of 114 tests passes, and the MCP server binary responds to JSON-RPC initialize on stdin/stdout. A Node.js v22.12.0 upgrade was applied (portable) and the cloakbrowser optional dependency was installed. Google Chrome is not... 4 分钟。

一个强大的 Model Context Protocol (MCP) 服务器，提供公共 Web 访问的一站式解决方案。

测试发现：The Bright Data MCP server installs and all 21 unit tests pass under Node.js v22.12.0; the server starts listening when API_TOKEN is set. 4 分钟。

Codex 计算机使用，与应用程序解耦，可在任何 harness 内使用。

测试发现：LCU 0.9.1 installed and verified extracting the official ChatGPT 26.915.31945 (CUA 0.0.16) app fixture, Python 435-test suite passes, adapter tests 88/103 pass, and lcu doctor confirms window listing and screenshot on virtual display. 7 分钟。

让你的 AI agent 获取登录和付费墙背后的源代码。

测试发现：Actionbook monorepo builds and all test suites pass (js-sdk 78/78, mcp 45/45, tools-ai-sdk 7/7, openclaw-plugin 39/39, cli 482/483), MCP server starts on stdio listing 5 tools, Rust CLI reports version 1.6.0, and json-ui renders reports, with the only gap being lsof not in the container and the remote API returning a... 13 分钟。

自托管网络搜索，适用于AI agents和SEO。Google, Bing, Yandex, Baidu, DuckDuckGo & Ecosia，在真实浏览器中渲染：排序结果和干净的页面Markdown。API, CLI, MCP

测试发现：OpenSERP v0.8.13 builds, all 12 unit test suites pass, and the HTTP server starts, serving /health and /ready endpoints on 127.0.0.1:7000. 8 分钟。

Scrapy 的 Playwright 集成

测试发现：scrapy-playwright 0.0.48 installed and builds; chromium handler works end-to-end (73 tests pass); firefox works standalone but 3 firefox test classes hang in pytest-twisted; webkit tests correctly skip on Linux; missing `file` system command patched in test. 61 分钟。

WebAI2API: 基于 Camoufox 的网页 AI 转 OpenAI API 工具，支持 LMArena/Gemini 等，多窗口并发与账号隔离。

测试发现：The WebAI2API server starts on port 3000 with a Camoufox browser instance, serving OpenAI-compatible /v1/models, WebUI, and /v1/cookies endpoints, all returning HTTP 200, though chat completions require real AI website credentials via the browser adapter and cannot produce output without logging in. 7 分钟。

Gemini Nexus 是一款面向浏览器场景的 AI 助手扩展，集成 Gemini Web、Gemini API 与 OpenAI 兼容接口，支持网页上下文、图像处理、工具调用和 MCP 浏览器控制。

测试发现：Gemini Nexus installs and builds successfully on Node 18 after downgrading jsdom to 26.x and knip to 5.x; all 1437 tests across 201 test files pass via npm run check; npm run package:extension produces a valid artifacts/chrome-extension directory; the Vite dev server serves the sidepanel, sandbox, and settings... 13 分钟。

为你的 AI agent 提供一个真正的浏览器，并让人参与其中。开源 MCP 原生浏览器 agent。

测试发现：Controller and client packages install in a venv; 1189 test suite tests pass via FastAPI TestClient with mocked browser node; Docker-based compose stack not available so full end-to-end with Playwright cannot run. 6 分钟。

gemini drawing MCP 与 skill 通过浏览器实现，可用于 openclaw 或任何支持 MCP 的 agent。

测试发现：Dependencies install cleanly, the MCP server starts and enumerates all 21 tools, the Atlas OpenAI-compatible provider works end-to-end against a local mock server, and the daemon starts with /health returning HTTP 200; the Gemini browser-driven tools could not be exercised because no Chrome/Chromium binary exists in... 9 分钟。

AI agent 控制 Chrome 的 CLI。零配置，与 agent 无关，久经考验。

测试发现：Surf v2.20.0 builds and all 81 test files pass (1177/1179 tests) using Node.js v22.12.0. The two test files that failed with SyntaxError were fixed by converting import-destructure patterns from require() to ESM import syntax. 19 分钟。

使用 agent-browser-cli 进行浏览器感知与控制。适用于标签页扫描/切换、页面 JS 执行、Cookie、CDP、contentSettings、截图、文件上传、下拉框点击、tmwd_cdp_bridge 初始化和 Web 工具故障排除。

测试发现：Daemon runs on default ports with config matched; with the mock extension connected, status/doctor/tabs/tabtree/scan/exec/open/close/lookup/profile-label/network/console/screenshot/save-pdf/install-skill all return ok:true and doctor reports all checks ok, while click/fill/send-keys on @e refs and mouse-click remain... 35 分钟。

本地法语语音助手：Claude 或 Ollama（离线）、Hue 智能家居、OBS、日程、浏览器、Twilio 通话、MCP 服务器。Python。

测试发现：Jarvis installs cleanly, its 181-test suite passes (4 Windows-only cases skipped), the Signal Matin pipeline emits valid A4 PDFs in all three density modes without overflow, and the OpenAI provider answers through a schema-correct local mock; only a physical microphone is missing. 36 分钟。

一项面向 AI agent 的技能：使用 SearXNG 搜索网络，使用 Camofox 浏览网页，使用 CloakBrowser 绕过保护。设计上防幻觉。自托管、免费、无限制。

测试发现：npm install succeeds, CloakBrowser (with Node 20, not the system Node 18) fetches live pages, the SearXNG and Camofox deterministic scripts pass their full command flows against schema-correct mocks, unit suites pass 28/28 and 12/12, and setup.sh/audit.sh complete without failed checks; real SearXNG/Camofox Docker... 33 分钟。

AI 说它完成了。Flyto2 展示了证据。

测试发现：The Python package installs, builds, and 4751 unit/module/contract tests pass. The CLI responds to help. Documentation and brand checks pass. The container lacks system ping and a newer Node for jsdom@29, both worked around. 54 分钟。

Browser Use Box：一个 24/7 运行的 Claude Code agent，用于 Playwright 风格的浏览器自动化，支持 Browser Use Cloud、Telegram 以及你拥有的任何设备上的真实浏览器。

测试发现：Python venv at /tmp/bux-venv with websockets, httpx, Pillow installed. All 84 agent tests pass (telegram_bot: 63, mini_app: 16, bux_miniapp_tunnel: 3, login_flow: 2). The project can't fully launch without BROWSER_USE_API_KEY, TG_BOT_TOKEN, and systemd, those are real third-party services, but the full test suite... 4 分钟。

平台无关的 AI agent 浏览器使用：通过 CLI + SDKs 实现真实的 Chrome 自动化，无锁定，极其简单。

测试发现：The repo installs and builds in this container; the Go CLI/MCP/SDKs run and answer requests through a standing mock of the Chrome extension, and every test suite in the repo passes after the crypto fix. 13 分钟。

在 Herdr 窗格内渲染一个真实的 Chromium 视图，并通过 CDP 驱动它。

测试发现：Project builds, typechecks clean, all 140 tests pass, CLI help displays, and test HTTP server starts on localhost. 33 分钟。

将你的 ChatGPT 或 Claude 账户变成一个完全可用的 OpenAI 兼容 API。无需 API 密钥。支持工具调用、视觉、文件附件和图像生成。

测试发现：CatGPT Gateway starts with PROVIDER=minimax, passes all 7 MiniMax provider tests, serves health/endpoints/models/chat completions on port 8000, and responds correctly to OpenAI SDK and curl requests. 6 分钟。

AI 驱动的登录自动化。使用 Claude 对登录页面进行分类，并使用 Playwright 与其交互。

测试发现：The Next.js app builds, launches, returns HTTP 200 on / and /api/chat, and responds to all API actions (start, submit, close, error handling) using local Playwright Chromium with a mock heuristic page classifier. 17 分钟。

Agent 技能，用于通过浏览器对 SaaS 产品进行逆向工程，将其转化为知识库，用于重建该产品：界面、规则、数据模型、设计 tokens 和生产障碍，且不触碰生产数据。

测试发现：All 32 automated tests pass under Node.js v22.14.0 and Playwright Chromium headless; all 4 CLI scripts exit 0 with correct usage and the end-to-end workflow (init, check, build-report, pack) produces valid output. 8 分钟。

pi 扩展，将 agent-browser 作为浏览器自动化的原生工具

测试发现：npm install, build, and doctor pass; 1137 of 1256 tests pass with fake-agent-browser fixtures; 79 pre-existing cold-reopen and daemon-boundary failures remain. 84 分钟。

一个 MCP 服务，允许 agent 直接操控真实 Chrome，支持页面扫描、CDP、截图与物理输入

测试发现：In a fresh .venv the project installs and builds, its CLI prints help/extension path/hermes config/doctor JSON, its MCP stdio server responds to initialize and tools/list with 19 tools, and against a mock Chrome bridge list_tabs, execute_js, scan_page, page screenshot, and desktop screenshot all returned real results... 28 分钟。

将你的电脑变成一台可以从手机操控的 AI agent 工作站。

测试发现：Opensteer 0.11.2 is installed as a uv global tool, all 100 unit tests pass, the CLI responds to --version (0.11.2), --help, and -c execution, and skills are installed. 1 分钟。

DeepSeek Harness 插件：为你的 agent 提供一个具有持久身份的浏览器，具备引擎级指纹伪装、无限免费本地配置文件、Android 设备模拟、可持久保留的通行密钥和住宅代理出口。

测试发现：The dsh-antibrow plugin package installs and packs cleanly; the smoke test passes with a mock MCP server implementing all 16 AntiBrow browser automation tools, verifying that the harness plugin bridge correctly discovers and routes all tools as mcp__antibrow__* entries. 11 分钟。

AI 编码工作站：Claude Code + Web UI + 8 个 AI CLI + 无头浏览器 + 50 多个工具

测试发现：All 285 Node.js unit tests, 5 Python notification tests, and the CLI persistence shell test pass. Repository structure is validated end to end with product facts contracts, security policy, documentation, and patch scripts all verified. 2 分钟。

让 Agent 直接操作真实 Chrome 的 MCP 服务，支持页面扫描、CDP、截图与物理输入

测试发现：agent-browser-mcp CLI and MCP server install, start, and respond to all 20 tools; desktop automation tools work against the Xvfb display; browser CDP tools await a Chrome extension connection. 6 分钟。

Eko（Eko Keeps Operating），使用自然语言构建生产级智能体工作流，eko.fellou.ai

测试发现：All 4 packages build successfully. 12 of 14 test suites pass (14 of 16 individual tests) in eko-core with a mock LLM server. Remaining failures: test/llm/utils.test.ts has a pre-existing unhandled rejection for an intentional timeout scenario; test/llm/mcp.test.ts requires an external MCP SSE server on localhost:8083... 27 分钟。

一个 macOS 应用，用于在 IDE 中运行 24/7 的 AI agents，具备记忆、定时任务、浏览器使用功能，并可访问 Whatsapp、Telegram、Slack。

测试发现：The dorabot backend compiles from source, its WebSocket gateway serves HTTP health at 200 on a Unix socket and handles RPC methods for auth, sessions, skills, agents, calendar RRULE scheduling, goals CRUD, projects/tasks, and provider detection against a real SQLite (WAL) database; the interactive CLI lists skills and... 29 分钟。

面向日常任务的浏览器 AI agent 的开源 benchmark。

测试发现：Python package clawbench-eval v0.10.0 installs and builds; all CLIs (clawbench, clawbench-run, clawbench-batch, clawbench-rescore, clawbench-analyze, clawbench-reproduce, clawbench-harbor-adapt, clawbench-edgebench-adapt, clawbench-edgebench-judge, clawbench-prorl-submit, clawbench-prorl-mock) are importable and... 3 分钟。

你的 AI 朋友就在你的浏览器中。

测试发现：npm install completed 771 packages with warnings about node version requirements; npm run build produces dist/ with manifest.json, background.js, sidepanel/panel.{js,css,html}, and content scripts; npm test passes 233 checks; integration tests pass 19; orchestrator harness passes 7; tsc typechecks without errors... 12 分钟。

功能强大的验证码求解器（Turnstile、cf_clearance、reCAPTCHA v3、AWS WAF），基于 FastAPI + Camoufox 构建。高成功率，支持代理，且对 VPS 友好。

测试发现：The FastAPI server runs on 0.0.0.0:8000, registers all 5 endpoints (/turnstile, /clearance, /aws-token, /recaptchaV3 GET+POST, /result), initializes a browser pool via camoufox, and correctly accepts task submissions and returns pollable results. 4 分钟。

开源 AI 同事各自拥有一台专属计算机：浏览器、文件和工具，所有操作都在执行前决定，并在执行后记录。可接入任意 AG-UI agent。

测试发现：bun install and all 3985 non-integration unit tests pass. The 69 failures are all infrastructure-gated (no PostgreSQL for integration tests, no Docker for compose tests, 2 pre-existing Mastra .env leakage failures). 49 分钟。

一个代理浏览器工具集：在你已登录的 Chrome、Edge、Arc、Brave 甚至 Firefox 中，以 AI 侧边面板和悬浮球的形式运行，由你已经在使用的 agent/订阅（Claude Code、Codex、Antigravity）驱动。免费、开源、无需 API 密钥。同时也是一个 MCP 服务器。

测试发现：无法验证；log 显示了它停在哪里。 8 分钟。