Testmu Browser Agent Public

测试转向——代理

AI原生浏览器自动化CLI和Chrome MCP服务器，内置于Go中。

  

与您的AI编码工具配合使用

与任何可以运行shell命令或通过MCP连接的代理配合使用。 自动安装所有插件 一个脚本。

______________________________________________________________________

68多个CLI命令。10个MCP工具。90多个行动。稳定 @ref 元素ID。单二进制。零依赖。

______________________________________________________________________

什么是testmu浏览器代理？

testmu-browser-agent是一个单一的Go二进制文件，它通过三个表面公开浏览器自动化：68+命令CLI、用于Claude Code的10工具MCP服务器和REST/SSE守护程序API-总共90多个操作。

• 68+命令行界面命令 --导航、单击、填充、截图、网络拦截、身份验证库、设备模拟、视频录制、HAR捕获等，所有这些都可以从终端完成
• 22项CDP行动 -通过MCP工具和REST API提供高级诊断、安全和性能操作
• 配备10个工具的MCP服务器 --直接插入Claude Code，这样AI代理就可以控制浏览器，而无需任何额外的设置
• 可访问性快照 @ref 鉴定 --DOM突变后仍能存活的令牌高效、稳定的元素引用
• LambdaTest云集成 --使用单个标志在真实的云浏览器上运行会话；无需基础设施
• AES-256-GCM加密 --会话状态在静止时使用用户提供的存储密钥进行加密
• 具有REST API和SSE的守护进程模式 --具有HTTP端点和服务器发送事件流的长期浏览器进程，用于实时事件监控

CLI / MCP / REST API
       |
  testmu-browser-agent (single Go binary)
       |
  Chrome (local) or LambdaTest (cloud)

______________________________________________________________________

快速开始

# Install
curl -sSL https://raw.githubusercontent.com/4DvAnCeBoY/testmu-browser-agent-public/main/scripts/install.sh | sh

# Open a page
testmu-browser-agent open https://example.com

# Take an accessibility snapshot (shows @ref IDs for all interactive elements)
testmu-browser-agent snapshot

# Click a button by its @ref ID
testmu-browser-agent click @e1

# Fill a form field
testmu-browser-agent fill @e5 "hello@example.com"

# Take a screenshot
testmu-browser-agent screenshot --output page.png

# Close the browser
testmu-browser-agent close

______________________________________________________________________

安装

安装脚本（macOS和Linux）

curl -sSL https://raw.githubusercontent.com/4DvAnCeBoY/testmu-browser-agent-public/main/scripts/install.sh | sh

该脚本检测您的操作系统和体系结构，从最新版本下载正确的二进制文件，并将其安装到 /usr/local/bin.

手动下载

从以下网址下载适用于您平台的预构建二进制文件 发布 页面：

使二进制文件可执行并将其移动到PATH上的一个目录中：

chmod +x testmu-browser-agent-darwin-arm64
mv testmu-browser-agent-darwin-arm64 /usr/local/bin/testmu-browser-agent

CLI命令

所有命令接受 全球旗帜.使用以下命令运行任何命令 --help 完全使用。

导航

testmu-browser-agent open           # Open a URL in a new browser page
testmu-browser-agent navigate       # Navigate current page to a URL
testmu-browser-agent goto           # Alias for navigate
testmu-browser-agent back                # Go back in browser history
testmu-browser-agent forward             # Go forward in browser history
testmu-browser-agent reload              # Reload the current page
testmu-browser-agent close               # Close the browser
testmu-browser-agent quit                # Alias for close
testmu-browser-agent exit                # Alias for close

交互

# Click
testmu-browser-agent click            # Click an element
testmu-browser-agent dblclick         # Double-click an element

# Forms
testmu-browser-agent fill       # Fill a form field with text
testmu-browser-agent type                     # Type into the focused element
testmu-browser-agent select    # Choose a dropdown option
testmu-browser-agent check            # Check a checkbox
testmu-browser-agent uncheck          # Uncheck a checkbox

# Keyboard
testmu-browser-agent press                     # Press a keyboard key (e.g. Enter, Tab)
testmu-browser-agent key                       # Alias for press
testmu-browser-agent keydown                   # Hold a key down
testmu-browser-agent keyup                     # Release a held key

# Pointer
testmu-browser-agent hover            # Hover over an element
testmu-browser-agent tap              # Tap (mobile-style)
testmu-browser-agent drag                 # Drag from one element to another
testmu-browser-agent scroll  [amount]    # Scroll the page (up/down/left/right)
testmu-browser-agent swipe  [distance]   # Swipe gesture
testmu-browser-agent mouse  [x] [y]         # Raw mouse operations (move, click, wheel)

# Files and focus
testmu-browser-agent upload     # Upload a file to a file input
testmu-browser-agent focus            # Focus an element

查询

# Accessibility snapshot — the recommended way to inspect pages
testmu-browser-agent snapshot                       # Full accessibility snapshot with @ref IDs
testmu-browser-agent snapshot --full                # Full tree including hidden elements
testmu-browser-agent snapshot --diff                # Show only what changed since last snapshot
testmu-browser-agent snapshot --max-length 4000     # Cap output length

# Get page content
testmu-browser-agent get text                       # Get visible text of the page
testmu-browser-agent get title                      # Get page title
testmu-browser-agent get url                        # Get current URL
testmu-browser-agent get text             # Get text of a specific element
testmu-browser-agent get attr       # Get an element attribute
testmu-browser-agent get count            # Count elements matching a selector
testmu-browser-agent get box              # Get bounding box of an element
testmu-browser-agent get styles           # Get computed styles of an element

# Find and inspect
testmu-browser-agent find                 # Find all elements matching a CSS selector
testmu-browser-agent find --role button             # Find by ARIA role
testmu-browser-agent find --text "Sign in"          # Find by visible text content
testmu-browser-agent find --label "Email"           # Find by aria-label or associated label
testmu-browser-agent find --placeholder "Search"    # Find by placeholder attribute
testmu-browser-agent find --alt "Logo"              # Find by alt attribute
testmu-browser-agent find --title "Close"           # Find by title attribute
testmu-browser-agent find --testid "submit-btn"     # Find by data-testid attribute
testmu-browser-agent find --role button --first     # Return only the first match
testmu-browser-agent find --role link --nth 3       # Return the 3rd match
testmu-browser-agent find --role button --last      # Return only the last match
testmu-browser-agent eval               # Evaluate JavaScript in the page
testmu-browser-agent inspect                        # Inspect page info and metadata

媒体

# Screenshots
testmu-browser-agent screenshot                     # Full-page screenshot (PNG)
testmu-browser-agent screenshot --output page.png   # Save to file
testmu-browser-agent screenshot --ref @e5        # Screenshot a single element
testmu-browser-agent screenshot --format jpeg --quality 90

# PDF and recording
testmu-browser-agent pdf output.pdf                 # Generate a PDF of the current page
testmu-browser-agent record start                   # Start screen recording
testmu-browser-agent record stop                    # Stop and save recording
testmu-browser-agent record restart                 # Restart recording

标签页

testmu-browser-agent tabs                           # List open tabs
testmu-browser-agent tab new                        # Open a new tab
testmu-browser-agent tab new https://example.com    # Open a new tab with a URL
testmu-browser-agent tab close                  # Close a tab by ID
testmu-browser-agent tab switch                 # Switch to a tab by ID
testmu-browser-agent window new                     # Open a new browser window
testmu-browser-agent frame                # Switch into an iframe

状态

# Save and load full browser state (cookies, storage, etc.)
testmu-browser-agent state save --name my-session
testmu-browser-agent state load --name my-session

# Cookies
testmu-browser-agent cookies get                    # List all cookies
testmu-browser-agent cookies set                    # Set a cookie
testmu-browser-agent cookies clear                  # Clear all cookies

# localStorage / sessionStorage
testmu-browser-agent storage get               # Get a storage value
testmu-browser-agent storage set        # Set a storage value
testmu-browser-agent storage clear                  # Clear all storage
testmu-browser-agent storage remove            # Remove a specific key
testmu-browser-agent storage get  --session    # Use sessionStorage

# Clipboard
testmu-browser-agent clipboard read                 # Read clipboard content
testmu-browser-agent clipboard write "text"         # Write to clipboard

等待

testmu-browser-agent wait --selector "#result"      # Wait for an element to appear
testmu-browser-agent wait --url "/dashboard"        # Wait for URL to match a pattern
testmu-browser-agent wait --text "Welcome"          # Wait for text to appear on page
testmu-browser-agent wait --load networkidle        # Wait for load state (domcontentloaded, load, networkidle)
testmu-browser-agent wait --fn "() => window.ready" # Wait for JS function to return truthy
testmu-browser-agent wait --download                # Wait for a download to complete
testmu-browser-agent wait --timeout 60000           # Set custom timeout in milliseconds (default: 30000)

网络和监控

testmu-browser-agent console                        # Get captured console messages
testmu-browser-agent console --clear                # Get and clear console messages
testmu-browser-agent errors                         # Get page JavaScript errors
testmu-browser-agent errors --clear                 # Get and clear errors
testmu-browser-agent dialog accept                  # Accept a browser dialog (alert/confirm)
testmu-browser-agent dialog dismiss                 # Dismiss a browser dialog
testmu-browser-agent highlight        # Visually highlight an element
testmu-browser-agent stream                         # Stream browser events (SSE)
testmu-browser-agent stream --filter "click"        # Filter events by type

开发者工具

testmu-browser-agent trace start                    # Start a Chrome DevTools trace
testmu-browser-agent trace stop                     # Stop tracing and save
testmu-browser-agent profiler start                 # Start the JS CPU profiler
testmu-browser-agent profiler stop                  # Stop and save profiler output
testmu-browser-agent batch ''        # Execute multiple commands atomically
testmu-browser-agent batch --bail '' # Stop batch on first error

元素查询

testmu-browser-agent is visible            # Check if element is visible
testmu-browser-agent is hidden             # Check if element is hidden
testmu-browser-agent is enabled            # Check if element is enabled
testmu-browser-agent is checked            # Check if checkbox is checked
testmu-browser-agent scrollintoview        # Scroll an element into the viewport

下载

testmu-browser-agent download                        # Enable download tracking and wait for next download
testmu-browser-agent download --dir /tmp/downloads   # Set download directory
testmu-browser-agent download --timeout 60000        # Custom timeout in milliseconds

网络侦听

testmu-browser-agent route "**/*.png" --abort        # Block all PNG requests
testmu-browser-agent route "/api/data" --body '{"mock":true}' --status 200  # Mock an API response
testmu-browser-agent route "/api/*" --header "X-Test:true"  # Add response headers
testmu-browser-agent unroute "/api/data"             # Remove a specific interception rule
testmu-browser-agent unroute                         # Remove all interception rules

认证保险库

testmu-browser-agent auth save --name mysite --url https://example.com/login --username user --password pass
testmu-browser-agent auth login --name mysite        # Auto-login using stored credentials
testmu-browser-agent auth list                       # List stored credentials (passwords masked)
testmu-browser-agent auth show --name mysite         # Show credential details
testmu-browser-agent auth delete --name mysite       # Delete a stored credential

差异

testmu-browser-agent diff snapshot                   # Diff accessibility tree against last snapshot
testmu-browser-agent diff url                   # Navigate to URL and diff before/after
testmu-browser-agent diff screenshot                 # Diff current screenshot against last stored

设备仿真

testmu-browser-agent geolocation 37.7749 -122.4194  # Override browser geolocation
testmu-browser-agent timezone America/New_York       # Override browser timezone
testmu-browser-agent locale fr-FR                    # Override browser locale
testmu-browser-agent permissions geolocation notifications  # Grant browser permissions
testmu-browser-agent offline                         # Enable offline mode
testmu-browser-agent offline --disable               # Restore network connectivity
testmu-browser-agent device-list                     # List available device profiles
testmu-browser-agent device-emulate "iPhone 15"      # Emulate a device (viewport, UA, scale)

内容注入

testmu-browser-agent addscript 'console.log("hi")'  # Evaluate JS in the current page
testmu-browser-agent addinitscript 'window.x = 1'   # Register JS that runs on every new document
testmu-browser-agent addstyle 'body { color: red }'  # Inject CSS into the current page
testmu-browser-agent expose myCallback               # Expose a function to page JS (calls forwarded as SSE events)

视频录制

testmu-browser-agent video start                     # Start recording page frames (screencast-based)
testmu-browser-agent video stop                      # Stop recording and save as GIF
testmu-browser-agent video stop --format frames      # Stop and save individual PNG frames
testmu-browser-agent video start --output-dir /tmp   # Set output directory for recordings
testmu-browser-agent video start --quality 90        # Set screencast quality (0-100)

HAR 捕获

testmu-browser-agent har start                       # Start capturing network traffic in HAR format
testmu-browser-agent har stop                        # Stop and return HAR data inline
testmu-browser-agent har stop --path traffic.har     # Stop and save HAR to file

网络请求检查

testmu-browser-agent requests                        # List all captured network requests
testmu-browser-agent requests --filter "api"         # Filter requests by URL pattern
testmu-browser-agent request-detail              # Get full details for a specific request
testmu-browser-agent response-body               # Get the response body for a specific request

流媒体

testmu-browser-agent stream-enable                   # Subscribe to CDP events via SSE
testmu-browser-agent stream-enable --events console,network  # Subscribe to specific event categories
testmu-browser-agent stream-disable                  # Unsubscribe from CDP events
testmu-browser-agent stream-status                   # Check if CDP event streaming is active
testmu-browser-agent screencast start                # Stream live page frames as base64 images
testmu-browser-agent screencast stop                 # Stop screencast

会话

同时运行多个独立的浏览器会话。每个会话都有自己的守护进程和浏览器实例。

# Default session (backward compatible — no flag needed)
testmu-browser-agent open https://example.com

# Named sessions
testmu-browser-agent --session work open https://github.com
testmu-browser-agent --session personal open https://gmail.com

# Check current session
testmu-browser-agent session

# List all active sessions
testmu-browser-agent session list

# JSON output
testmu-browser-agent session list --output json

每个会话都会在中创建一个单独的守护进程，该守护进程具有自己的Unix套接字 ~/.testmu-browser-agent/The --socket 标志覆盖了基于会话的套接字解析，以实现向后兼容性。

CDP/高级浏览器控制（仅限MCP和REST API）

用于高级诊断、安全测试和性能分析的低级Chrome DevTools协议操作。这些操作可通过 MCP工具 (browser_config, browser_state, browser_network, browser_devtools, browser_query)以及 REST API --而不是作为独立的CLI命令。

维护

testmu-browser-agent install                         # Download and install Chrome for Testing
testmu-browser-agent install --dest /opt/chrome      # Install to a custom directory
testmu-browser-agent upgrade                         # Self-update to the latest release

配置

testmu-browser-agent set viewport 1920x1080         # Set browser viewport
testmu-browser-agent set useragent "MyAgent/1.0"    # Override the user agent
testmu-browser-agent set geolocation "37.77,-122.4" # Set geolocation
testmu-browser-agent set offline true               # Enable offline mode
testmu-browser-agent connect               # Connect to a remote CDP endpoint
testmu-browser-agent device list                    # List emulatable device profiles

______________________________________________________________________

MCP服务器（克劳德代码集成）

testmu浏览器代理附带了一个模型上下文协议服务器，Claude Code可以在对话期间使用它来控制真正的Chrome浏览器。MCP服务器公开了10个分组工具，并通过stdio进行通信，不需要网络端口。

设置

将以下内容添加到您的Claude Code设置文件中(.claude/settings.json 或 ~/.claude/settings.json):

{
  "mcpServers": {
    "testmu-browser-agent": {
      "command": "testmu-browser-agent",
      "args": ["mcp"],
      "env": {}
    }
  }
}

对于无头模式（在CI中或当您不想要可见的浏览器窗口时有用）：

{
  "mcpServers": {
    "testmu-browser-agent": {
      "command": "testmu-browser-agent",
      "args": ["mcp", "--headless"]
    }
  }
}

保存后，重新启动Claude Code。浏览器工具将在您的下一次对话中可用。

可用工具

有关完整的集成指南，请参阅 docs/guides/mcp-integration.md.

______________________________________________________________________

LambdaTest云

在LambdaTest的云基础设施上运行浏览器会话，而不是在本地安装Chrome。这对于CI管道、Windows上的跨浏览器测试以及避免本地浏览器设置非常有用。

环境变量

export LT_USERNAME="your-lambdatest-username"
export LT_ACCESS_KEY="your-lambdatest-access-key"

带LambdaTest的CLI

testmu-browser-agent --provider lambdatest open https://example.com
testmu-browser-agent --provider lambdatest snapshot
testmu-browser-agent --provider lambdatest screenshot --output result.png

守护进程与LambdaTest

testmu-browser-agent serve --provider lambdatest --port 9222

带LambdaTest的MCP

{
  "mcpServers": {
    "testmu-browser-agent": {
      "command": "testmu-browser-agent",
      "args": ["mcp", "--provider", "lambdatest"],
      "env": {
        "LT_USERNAME": "your-username",
        "LT_ACCESS_KEY": "your-access-key"
      }
    }
  }
}

LambdaTest会话会自动在LambdaTest仪表板中捕获视频、控制台日志和网络流量。提供者每60秒发送一次keepalive ping，以保持长时间运行的会话处于活动状态，并在会话关闭时将测试标记为通过或失败。

有关完整的LambdaTest设置指南，请参阅 docs/guides/lambdatest.md.

______________________________________________________________________

Appium移动测试

通过Appium服务器在真实的iOS和Android设备上运行浏览器会话。这对于移动网络测试和响应式验证非常有用。

环境变量

export APPIUM_URL="http://localhost:4723"   # Appium server URL (or set via --appium-url)

带Appium的CLI

testmu-browser-agent --provider appium --platform android open https://example.com
testmu-browser-agent --provider appium --platform ios snapshot
testmu-browser-agent --provider appium --appium-url http://hub:4723 --platform android screenshot --output mobile.png

带Appium的MCP

{
  "mcpServers": {
    "testmu-browser-agent": {
      "command": "testmu-browser-agent",
      "args": ["mcp", "--provider", "appium", "--platform", "android"],
      "env": {
        "APPIUM_URL": "http://localhost:4723"
      }
    }
  }
}

______________________________________________________________________

守护程序模式

Daemon模式启动一个长期运行的浏览器进程，并通过HTTP和Unix域套接字公开它。CLI命令会自动连接到正在运行的守护进程，因此您可以在许多命令中重用单个浏览器会话，而无需每次启动新浏览器的启动开销。

启动守护进程

testmu-browser-agent serve                          # HTTP on port 9222 (default)
testmu-browser-agent serve --port 9000              # Custom port
testmu-browser-agent serve --socket /tmp/tmu.sock   # Unix socket
testmu-browser-agent serve --headless               # Headless mode
testmu-browser-agent serve --provider lambdatest    # Cloud browser

REST API

每个CLI操作都可以作为HTTP端点使用：

POST /{action}   — JSON body with action parameters
GET  /{action}   — query string parameters
GET  /events     — Server-Sent Events stream

示例：

# Navigate
curl -X POST http://localhost:9222/open \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

# Take a snapshot
curl http://localhost:9222/snapshot

# Get page title
curl "http://localhost:9222/get?type=title"

# Subscribe to browser events
curl -N http://localhost:9222/events

SSE事件流

这 /events 端点使用服务器发送事件实时流式传输所有浏览器事件。客户可以订阅任何支持SSE的库或 curl -N.

curl -N http://localhost:9222/events
# data: {"type":"navigate","url":"https://example.com","timestamp":"..."}
# data: {"type":"click","ref":"@e12","timestamp":"..."}

空闲超时

默认情况下，守护进程在5分钟不活动后自行关闭。每个传入请求都会重置计时器。您可以在 serve 命令。

______________________________________________________________________

辅助功能快照

snapshot 是理解页面内容的主要方式。它不是返回原始HTML，而是生成一个紧凑、结构化的可访问性树，其中每个交互元素都被分配了一个稳定的 @ref 身份证件。

输出示例

[1] Page: Example Domain
  [2] heading: Example Domain @e1
  [3] paragraph: This domain is for use in illustrative examples...
  [4] link: More information... @e2
    href: https://www.iana.org/domains/reserved

然后，您可以直接引用任何元素：

testmu-browser-agent click @e2
testmu-browser-agent fill @e5 "search query"

为什么是快照而不是截图？

• 稳定的引用 — @ref ID在页面重新呈现和微小的DOM更改中幸存下来
• 全树模式 — --full 当您需要完整图片时，包含非交互式和隐藏元素
• 困难 — --diff 仅显示自上次快照以来发生的更改，非常适合验证操作是否具有预期效果
• 代币高效 --可访问性树比HTML或图像数据小得多，使其适用于LLM上下文窗口

______________________________________________________________________

全球旗帜

这些标志适用于每个命令：

______________________________________________________________________

环境变量

______________________________________________________________________

码头工人

在包含Chrome的Docker容器中构建并运行testmu浏览器代理：

# Build the image
docker build -t testmu-browser-agent:latest .

# Run a one-off command
docker run --rm testmu-browser-agent:latest open https://example.com

# Start the daemon and expose the REST API
docker run -p 9222:9222 testmu-browser-agent:latest serve --headless

# Run with LambdaTest
docker run --rm \
  -e LT_USERNAME=your-username \
  -e LT_ACCESS_KEY=your-access-key \
  testmu-browser-agent:latest --provider lambdatest open https://example.com

______________________________________________________________________

例子

______________________________________________________________________

文档

______________________________________________________________________

基准测试

运行基准测试套件来测量您计算机上的快照生成、操作调度和SSE吞吐量：

make benchmark

结果和历史数据见 基准测试/.