WebDriverIO MCP Server

A Model Context Protocol (MCP) server that enables AI assistants to interact with web browsers, local Electron applications, and mobile applications using WebdriverIO. Automate Chrome, Firefox, Edge, Safari, Electron, iOS, and Android through a unified interface.

Installation

Add the following configuration to your MCP client settings:

Standard config (works in most clients):

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), %APPDATA%\Claude\claude_desktop_config.json (Windows), or ~/.config/Claude/claude_desktop_config.json (Linux):

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

bashCopy code
claude mcp add wdio-mcp -- npx -y @wdio/mcp@latest

Add to your VS Code settings.json or cline_mcp_settings.json file:

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

Go to Cursor Settings → MCP → Add new MCP Server, or create .cursor/mcp.json:

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

Use the Codex CLI:

bashCopy code
codex mcp add wdio-mcp npx "@wdio/mcp@latest"

Or edit ~/.codex/config.toml:

tomlCopy code
[mcp_servers.wdio-mcp]
command = "npx"
args = ["@wdio/mcp@latest"]

Go to Advanced settings → Extensions → Add custom extension, or run:

bashCopy code
goose configure

Or edit ~/.config/goose/config.yaml:

yamlCopy code
extensions:
  wdio-mcp:
    name: WebDriverIO MCP
    cmd: npx
    args: [ -y, "@wdio/mcp@latest" ]
    enabled: true
    type: stdio

Edit ~/.codeium/windsurf/mcp_config.json:

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

Edit Zed settings (~/.config/zed/settings.json):

jsonCopy code
{
  "context_servers": {
    "wdio-mcp": {
      "source": "custom",
      "command": "npx",
      "args": [
        "-y",
        "@wdio/mcp@latest"
      ]
    }
  }
}

bashCopy code
code --add-mcp '{"name":"wdio-mcp","command":"npx","args":["-y","@wdio/mcp@latest"]}'

⚠️ Restart Required: After adding the configuration, fully restart your MCP client to apply the changes.

Option 2: Global Installation

If you prefer to install globally:

bashCopy code
npm install -g @wdio/mcp

Then use wdio-mcp as the command:

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "wdio-mcp"
    }
  }
}

📖 Need help? Follow the MCP install guide.

HTTP Transport (for non-subprocess clients)

By default the server uses stdio (subprocess) transport. For clients that cannot launch subprocesses (e.g. llama.cpp, OpenAI Codex secure mode), enable HTTP transport:

bashCopy code
npx @wdio/mcp --http --port 3000

Then point your MCP client at http://localhost:3000/mcp.

Prerequisites For Mobile App Automation

• Appium Server: Install globally with npm install -g appium
• Platform Drivers:
  • iOS: appium driver install xcuitest (requires Xcode on macOS)
  • Android: appium driver install uiautomator2 (requires Android Studio)
• Devices/Emulators:
  • iOS Simulator (macOS) or physical device
  • Android Emulator or physical device
• For iOS Real Devices: You'll need the device's UDID (Unique Device Identifier)
  • Find UDID on macOS: Connect device → Open Finder → Select device → Click device name/model to reveal UDID
  • Find UDID on Windows: Connect device → iTunes or Apple Devices app → Click device icon → Click "Serial Number" to reveal UDID
  • Xcode method: Window → Devices and Simulators → Select device → UDID shown as "Identifier"

Start the Appium server before using mobile features:

bashCopy code
appium
# Server runs at http://127.0.0.1:4723 by default

Cloud Providers

Run browser and mobile app tests on cloud real devices and browsers without any local setup. Currently supports BrowserStack, Sauce Labs, LambdaTest, TestingBot, and Digital.ai Testing.

Prerequisites

Set your provider credentials as environment variables or in your MCP client config:

bashCopy code
export BROWSERSTACK_USERNAME=your_username
export BROWSERSTACK_ACCESS_KEY=your_access_key

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": ["-y", "@wdio/mcp@latest"],
      "env": {
        "BROWSERSTACK_USERNAME": "your_username",
        "BROWSERSTACK_ACCESS_KEY": "your_access_key"
      }
    }
  }
}

bashCopy code
export SAUCE_USERNAME=your_username
export SAUCE_ACCESS_KEY=your_access_key

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": ["-y", "@wdio/mcp@latest"],
      "env": {
        "SAUCE_USERNAME": "your_username",
        "SAUCE_ACCESS_KEY": "your_access_key"
      }
    }
  }
}

| SAUCE_USERNAME | Sauce Labs username (required) | | SAUCE_ACCESS_KEY | Sauce Labs access key (required) |

The data center is set per-session via the region parameter in start_session (defaults to eu-central-1).

bashCopy code
export TESTMU_USERNAME=your_username
export TESTMU_ACCESS_KEY=your_access_key

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": ["-y", "@wdio/mcp@latest"],
      "env": {
        "TESTMU_USERNAME": "your_username",
        "TESTMU_ACCESS_KEY": "your_access_key"
      }
    }
  }
}

| TESTMU_USERNAME | LambdaTest username (required) | | TESTMU_ACCESS_KEY | LambdaTest access key (required) |

bashCopy code
export TESTINGBOT_KEY=your_key
export TESTINGBOT_SECRET=your_secret

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": ["-y", "@wdio/mcp@latest"],
      "env": {
        "TESTINGBOT_KEY": "your_key",
        "TESTINGBOT_SECRET": "your_secret"
      }
    }
  }
}

| TESTINGBOT_KEY | TestingBot key (required) | | TESTINGBOT_SECRET | TestingBot secret (required) |

bashCopy code
export DIGITALAI_CLOUD_URL=https://your-cloud.example.com
export DIGITALAI_ACCESS_KEY=your_access_key

jsonCopy code
{
  "mcpServers": {
    "wdio-mcp": {
      "command": "npx",
      "args": ["-y", "@wdio/mcp@latest"],
      "env": {
        "DIGITALAI_CLOUD_URL": "https://your-cloud.example.com",
        "DIGITALAI_ACCESS_KEY": "your_access_key"
      }
    }
  }
}

| DIGITALAI_CLOUD_URL | Digital.ai cloud host, e.g. https://your-cloud.example.com (required) | | DIGITALAI_ACCESS_KEY | Digital.ai access key (required) |

The access key is sent via the digitalai:options capability (mobile) or the flat digitalai:accessKey capability (web).

Report pass/fail status: WebdriverIO defaults to the BiDi protocol, over which Digital.ai's cloud cannot observe command failures — so reports default to "Passed". To make the cloud reflect actual pass/fail, opt in to classic WebDriver per session:

jsCopy code
start_session({
  provider: 'digitalai', platform: 'browser', browser: 'chrome', os: 'Windows 10',
  capabilities: { 'wdio:enforceWebDriverClassic': true }
})

(Pure client-side assertion failures still report as "Passed" — only failures that reach the cloud as WebDriver command errors are detected.)

Mobile (Appium): configure your Digital.ai project for Appium-server execution and pick its default Appium version via the project's "Manage default Appium server version" setting — the version is chosen at the project level (and tracks the versions your cloud supports), so this MCP does not pin one. See Appium Server Test Execution.

Browser Sessions

Run a browser on a specific OS/version combination:

javascriptCopy code
// BrowserStack
start_session({
    provider: 'browserstack',
    platform: 'browser',
    browser: 'chrome',           // chrome | firefox | edge | safari
    browserVersion: 'latest',    // default: latest
    os: 'Windows',               // e.g. "Windows", "OS X"
    osVersion: '11',             // e.g. "11", "Sequoia"
    reporting: {
        project: 'My Project',
        build: 'v1.2.0',
        session: 'Login flow'
    }
})

// Sauce Labs
start_session({
    provider: 'saucelabs',
    platform: 'browser',
    browser: 'chrome',
    os: 'Windows',               // combined with osVersion → platformName
    osVersion: '11',             // e.g. "11", "15" (numbered Mac naming)
    region: 'eu-central-1',      // default: eu-central-1
    reporting: {
        build: 'v1.2.0',
        session: 'Login flow'
    }
})

// LambdaTest
start_session({
    provider: 'testmu',
    platform: 'browser',
    browser: 'chrome',
    os: 'Windows',               // combined with osVersion → platformName
    osVersion: '11',             // e.g. "11", "Sequoia" (optional)
    reporting: {
        project: 'My Project',
        build: 'v1.2.0',
        session: 'Login flow'
    }
})

// TestingBot
start_session({
    provider: 'testingbot',
    platform: 'browser',
    browser: 'chrome',
    os: 'Windows',               // combined with osVersion → platformName (default: Windows 11)
    osVersion: '11',
    reporting: {
        build: 'v1.2.0',
        session: 'Login flow'
    }
})

// Digital.ai
start_session({
    provider: 'digitalai',
    platform: 'browser',
    browser: 'chrome',
    os: 'Windows',               // combined with osVersion → digitalai:osName (optional)
    osVersion: '11',
    reporting: {
        session: 'Login flow'    // → digitalai:testName (flat capability, not nested)
    }
})

Provider-specific os / osVersion behavior:

• BrowserStack — os and osVersion map to separate bstack:options.os / bstack:options.osVersion fields.
• Sauce Labs / LambdaTest / TestingBot — os and osVersion are combined into the W3C platformName capability (e.g., os: 'Windows' + osVersion: '11' → platformName: 'Windows 11'). These providers use platformName values like "Windows 11", "MacOS Sequoia", or "Linux". TestingBot defaults to Windows 11 when os is omitted.
• Digital.ai — os and osVersion are combined into the flat digitalai:osName capability (NOT platformName), e.g. os: 'Windows' + osVersion: '11' → digitalai:osName: 'Windows 11'.

Mobile App Sessions

Test on cloud real devices. First upload your app (or use an existing app URL):

javascriptCopy code
// BrowserStack: returns bs:// URL
upload_app({ provider: 'browserstack', path: '/path/to/app.apk' })

// Sauce Labs: returns storage:filename= reference
upload_app({ provider: 'saucelabs', path: '/path/to/app.apk' })

// LambdaTest: returns lt:// URL
upload_app({ provider: 'testmu', path: '/path/to/app.apk' })

// TestingBot: returns tb:// URL
upload_app({ provider: 'testingbot', path: '/path/to/app.apk' })

// Digital.ai: returns cloud:<package-or-bundle> reference
upload_app({ provider: 'digitalai', path: '/path/to/app.apk' })

// Start a session
start_session({
    provider: 'browserstack',
    platform: 'android',
    app: 'bs://abc123...',
    deviceName: 'Samsung Galaxy S23',
    platformVersion: '13.0'
})

// Sauce Labs native app
start_session({
    provider: 'saucelabs',
    platform: 'android',
    app: 'storage:filename=myapp.apk',
    deviceName: 'Samsung.*',
    platformVersion: '16'
})

// LambdaTest native app
start_session({
    provider: 'testmu',
    platform: 'android',
    app: 'lt://abc123...',
    deviceName: 'Pixel 7',
    platformVersion: '13'
})

// TestingBot native app
start_session({
    provider: 'testingbot',
    platform: 'android',
    app: 'tb://abc123...',
    deviceName: 'Pixel 7',
    platformVersion: '13'
})

// Digital.ai native app — devices are selected via a deviceQuery
start_session({
    provider: 'digitalai',
    platform: 'android',
    app: 'cloud:com.example.app',
    deviceQuery: "@os='android' and @version='14' and @name='.*Pixel.*'"
    // or omit deviceQuery and pass deviceName / platformVersion to build one
})

Mobile Browser Sessions

Run a browser on a cloud mobile device — real device or emulator/simulator — without uploading an app:

javascriptCopy code
// BrowserStack — Chrome on Android emulator
start_session({
    provider: 'browserstack',
    platform: 'android',
    browser: 'chrome',
    deviceName: 'Google Pixel 7',
    platformVersion: '13'
})

// Sauce Labs — Safari on iOS simulator
start_session({
    provider: 'saucelabs',
    platform: 'ios',
    browser: 'safari',
    deviceName: 'iPhone 15',
    platformVersion: '18',
    region: 'eu-central-1'
})

// LambdaTest — Chrome on Android emulator
start_session({
    provider: 'testmu',
    platform: 'android',
    browser: 'chrome',
    deviceName: 'Pixel 7',
    platformVersion: '13'
})

// TestingBot — Chrome on Android emulator
start_session({
    provider: 'testingbot',
    platform: 'android',
    browser: 'chrome',
    deviceName: 'Pixel 7',
    platformVersion: '13'
})

// Digital.ai — Chrome on an Android device (real or emulator; selected via a deviceQuery)
start_session({
    provider: 'digitalai',
    platform: 'android',
    browser: 'chrome',
    deviceName: 'Pixel 7',
    platformVersion: '13'
    // or omit deviceName / platformVersion and pass deviceQuery directly, e.g.
    // deviceQuery: "@os='android' and @emulator='true'" to force an emulator
})

Note: Mobile browser sessions do not require app, appPath, or noReset. The provider launches a browser directly on the selected device — real or emulator/simulator.

Use list_apps to see previously uploaded apps:

javascriptCopy code
list_apps({ provider: 'browserstack' })
list_apps({ provider: 'saucelabs', sortBy: 'app_name' })
list_apps({ provider: 'testmu' })
list_apps({ provider: 'testingbot' })
list_apps({ provider: 'digitalai' })
list_apps({ provider: 'browserstack', organizationWide: true })

Local Tunnel

To test against URLs that are only accessible on your local machine or internal network, enable a local tunnel:

javascriptCopy code
// Auto-start tunnel (provider manages lifecycle)
start_session({
    provider: 'saucelabs',
    platform: 'browser',
    tunnel: true                  // auto-starts tunnel before session
})

// Use an already-running tunnel
start_session({
    provider: 'saucelabs',
    platform: 'browser',
    tunnel: 'external'            // uses existing tunnel
})

The tunnel parameter replaces the deprecated browserstackLocal, saucelabsLocal, and testmuLocal params. Set it to true to auto-start the tunnel (stopped automatically after the session), or 'external' to use a tunnel already running on your machine.

Note: With tunnel: true the provider downloads and manages the tunnel binary for you. For tunnel: 'external' you run it yourself — the wdio://saucelabs/local-binary, wdio://testmu/local-binary, and wdio://testingbot/local-binary resources provide download URLs and setup instructions. The TestingBot Tunnel is a single cross-platform Java JAR (requires Java 11+) rather than a per-platform binary.

Reporting Labels

All session types support reporting labels that appear in the provider dashboard:

Cloud Provider Tools

Both tools require a provider parameter ('browserstack', 'saucelabs', 'testmu', 'testingbot', or 'digitalai').

Features

Browser Automation

• Session Management: Start and close browser sessions (Chrome, Firefox, Edge, Safari) with headless/headed modes
• Navigation & Interaction: Navigate URLs, click elements, fill forms, and retrieve content
• Page Analysis: Get visible elements, accessibility trees, take screenshots
• Cookie Management: Get, set, and delete cookies
• Scrolling: Smooth scrolling with configurable distances
• Attach to running Chrome: Connect to an existing Chrome window via --remote-debugging-port — ideal for testing authenticated or pre-configured sessions
• Connect to existing WebDriver endpoints: Reuse an already-running Selenium-compatible WebDriver endpoint, such as a framework-managed browser or a desktop webview automation bridge (like Tauri)
• Device emulation: Apply mobile/tablet presets (iPhone 15, Pixel 7, etc.) to simulate responsive layouts without a physical device
• Session Recording: All tool calls are automatically recorded and exportable as runnable WebdriverIO JS

Mobile App Automation (iOS/Android)

• Native App Testing: Test iOS (.app/.ipa) and Android (.apk) apps via Appium
• Touch Gestures: Tap, swipe, long-press, drag-and-drop
• App Lifecycle: Launch, background, terminate, check app state
• Context Switching: Seamlessly switch between native and webview contexts for hybrid apps
• Device Control: Rotate, lock/unlock, geolocation, keyboard control, notifications
• Cross-Platform Selectors: Accessibility IDs, XPath, UiAutomator (Android), Predicates (iOS)

Available Tools

Session Management

Navigation & Page Interaction (Web & Mobile)

Element Interaction (Web & Mobile)

Cookie Management (Web)

Mobile Gestures (iOS/Android)

Context Switching (Hybrid Apps)

Device Control (iOS/Android)

MCP Resources (read-only, no tool call needed)

Usage Examples

Real-World Test Cases

Example 1: Testing Demo Android App (Book Scanning)

Copy code
Test the Demo Android app at C:\Users\demo-liveApiGbRegionNonMinifiedRelease-3018788.apk on emulator-5554:
1. Start the app with auto-grant permissions
2. Get visible elements on the onboarding screen
3. Tap "Skip" to bypass onboarding
4. Verify main screen loads
5. Take a screenshot

Example 2: Testing World of Books E-commerce Site

Copy code
You are a Testing expert, and want to assess the basic workflows of worldofbooks.com:
- Open World of Books (accept all cookies)
- Get visible elements to see navigation structure
- Search for a fiction book
- Choose one and validate if there are NEW and used book options
- Report your findings at the end

Browser Automation

Basic web testing prompt:

Copy code
You are a Testing expert, and want to assess the basic workflows of a web application:
- Open World of Books (accept all cookies)
- Search for a fiction book
- Choose one and validate if there are NEW and used book options
- Report your findings at the end

Browser configuration options:

javascriptCopy code
// Default settings (headed mode, 1280x1080)
start_session({platform: 'browser'})

// Firefox
start_session({platform: 'browser', browser: 'firefox'})

// Edge
start_session({platform: 'browser', browser: 'edge'})

// Safari (headed only; requires macOS)
start_session({platform: 'browser', browser: 'safari'})

// Headless mode
start_session({platform: 'browser', headless: true})

// Custom dimensions
start_session({platform: 'browser', windowWidth: 1920, windowHeight: 1080})

// Pass custom capabilities (e.g. Chrome extensions, profile, prefs)
start_session({
    platform: 'browser',
    headless: false,
    capabilities: {
        'goog:chromeOptions': {
            args: ['--user-data-dir=/tmp/wdio-mcp-profile', '--load-extension=/path/to/unpacked-extension']
        }
    }
})

Electron applications

Electron support is local-only and uses the official @wdio/electron-service standalone lifecycle. It requires Node.js 22.12 or newer. Put service options such as appBinaryPath, appEntryPoint, and appArgs in capabilities['wdio:electronServiceOptions']; use top-level electronRootDir for the service's Electron Builder/Electron Forge discovery. When testing a binary outside the project, set browserVersion to the Electron version so the service can select a compatible Chromedriver.

javascriptCopy code
start_session({
  platform: 'electron',
  browserVersion: '33.2.1',
  capabilities: {
    'wdio:electronServiceOptions': {
      appBinaryPath: '/path/to/MyApp.app/Contents/MacOS/MyApp',
      appArgs: ['--disable-gpu']
    }
  }
})

// Privileged: this code runs in the Electron main process, not the renderer.
execute_electron_script({ script: 'return electron.app.getName()' })

Existing browser DOM tools work against the Electron renderer. close_session always tears down MCP-managed Electron sessions; detach: true is intentionally unsupported. Main/renderer log capture can be enabled with captureMainProcessLogs or captureRendererLogs plus logDir. Electron function mocks are available through mock, get_mock_calls, and manage_mock.

jsCopy code
mock({
  mockType: 'electron', apiName: 'dialog', funcName: 'showOpenDialog',
  behavior: 'mockResolvedValue', value: { canceled: false, filePaths: ['/tmp/example.txt'] }
})
// Interact with the renderer to open the application's file picker, then inspect its calls.
get_mock_calls({ mockType: 'electron', apiName: 'dialog', funcName: 'showOpenDialog' })
manage_mock({ mockType: 'electron', apiName: 'dialog', funcName: 'showOpenDialog', action: 'restore' })

mockType accepts 'electron' or 'browser'. WebDriver browser sessions default to 'browser' when omitted. Electron sessions require an explicit selection because they can target both types of mocks. Use mockType: 'electron' with apiName and funcName for main-process API functions in an active Electron session. Browser mocking is not implemented yet and returns a clear error for either runtime. iOS/Android Appium sessions do not support mocking.

behavior defaults to mockReturnValue; mockResolvedValue and mockRejectedValue support async APIs. Each has a Once variant for queued responses. Repeated configuration preserves the existing mock and call history. Values must be JSON; omit value for undefined. clear removes call history, reset also removes behavior and queued responses, and restore reinstates the original function. Handles belong to the active browser session and cannot be reused after it closes or is replaced. These tools support individual API functions; class mocks and arbitrary mock implementations are not exposed. All three tools participate in tracing and generated replay.

To trigger an app deeplink, explicitly configure its URI scheme when starting the Electron session. The scheme has no colon and only URLs with that exact scheme can be dispatched:

javascriptCopy code
start_session({
  platform: 'electron',
  electronDeeplinkScheme: 'myapp',
  capabilities: {
    'wdio:electronServiceOptions': {
      appBinaryPath: '/path/to/MyApp.app/Contents/MacOS/MyApp'
    }
  }
})
trigger_electron_deeplink({ url: 'myapp://open/item' })

Attach to a running Chrome instance:

Copy code
// First, launch Chrome with remote debugging enabled:
//
//   macOS (must quit Chrome first — open -a ignores args if Chrome is already running):
//     pkill -x "Google Chrome" && sleep 1
//     /Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome \
//       --remote-debugging-port=9222 \
//       --user-data-dir=/tmp/chrome-debug &
//
//   Linux:
//     google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-debug &
//
//   Verify it's ready: curl http://localhost:9222/json/version
start_session({attach: true})
start_session({attach: true, port: 9333})
start_session({attach: true, port: 9222, navigationUrl: 'https://app.example.com'})

Connect to an existing WebDriver endpoint:

Use provider: 'external' when another process already owns the browser or webview lifecycle and exposes a W3C WebDriver endpoint. This is useful for Selenium Grid sessions, externally managed browser drivers, or desktop apps like Tauri apps that embed a webview and expose WebDriver separately. The MCP server connects to the endpoint; it does not launch or stop the target application, start tunnels, or manage framework-specific setup.

javascriptCopy code
// Defaults to http://127.0.0.1:4445/ and browserName: 'chrome'
start_session({provider: 'external', platform: 'browser'})

// Custom WebDriver endpoint and capabilities
start_session({
    provider: 'external',
    platform: 'browser',
    webdriverConfig: {
        protocol: 'http',
        hostname: '127.0.0.1',
        port: 4445,
        path: '/'
    },
    capabilities: {
        browserName: 'tauri'
    }
})

For desktop webview apps such as Tauri, first start the app and its WebDriver bridge outside of this MCP server, then pass the endpoint and the required capabilities. For example, a Tauri WebDriver bridge may require capabilities: {browserName: 'tauri'}.

Attach to an existing WebDriver or Appium session:

Use attach_session when the session has already been created by another process. The MCP reuses the selected provider's endpoint and credentials, registers the appropriate browser/mobile command set locally, and does not issue a new-session request. Attached sessions are externally managed: close_session() detaches by default, while close_session({detach: false}) explicitly terminates the remote session.

javascriptCopy code
// Existing BrowserStack App Automate session
attach_session({
    provider: 'browserstack',
    platform: 'ios',
    sessionId: 'existing-browserstack-session-id',
    capabilities: {
        'appium:deviceName': 'iPhone 15',
        'appium:automationName': 'XCUITest'
    }
})

// Existing session on a local Appium server
attach_session({
    provider: 'local',
    platform: 'android',
    sessionId: 'existing-appium-session-id',
    appiumConfig: {
        protocol: 'http',
        host: '127.0.0.1',
        port: 4723,
        path: '/'
    }
})

// Existing mobile session on a custom W3C WebDriver endpoint
attach_session({
    provider: 'external',
    platform: 'ios',
    sessionId: 'existing-grid-session-id',
    webdriverConfig: {
        protocol: 'https',
        hostname: 'grid.example.com',
        port: 443,
        path: '/wd/hub'
    }
})

An existing cloud session must keep using the tunnel it was created with. attach_session never starts or stops a tunnel, so keep the original tunnel process alive for as long as the session needs it.

Device emulation (requires BiDi session):

Copy code
// Device emulation (requires BiDi session)
start_session({capabilities: {webSocketUrl: true}})
emulate_device()                         // list available presets
emulate_device({device: 'iPhone 15'})    // activate emulation
emulate_device({device: 'Pixel 7'})      // switch device
emulate_device({device: 'reset'})        // restore desktop defaults

Web extensions (requires BiDi session):

javascriptCopy code
start_session({platform: 'browser', browser: 'chrome', capabilities: {webSocketUrl: true}})

open_web_extension({
    extensionData: {type: 'path', path: '/path/to/unpacked-extension'},
    path: 'options.html'
})

// Drive the extension UI with the normal page tools.
get_elements()
click_element({selector: '#save'})

// For remote/cloud sessions, send a packaged extension archive as base64.
open_web_extension({
    extensionData: {type: 'base64', value: '<base64-encoded-zip>'},
    path: 'options.html'
})

Mobile App Automation

Testing an iOS app on simulator:

Copy code
Test my iOS app located at /path/to/MyApp.app on iPhone 15 Pro simulator:
1. Start the app session
2. Tap the login button
3. Enter "testuser" in the username field
4. Take a screenshot of the home screen
5. Close the session

Preserving app state between sessions:

Copy code
Test my Android app without resetting data:
1. Start app session with noReset: true and fullReset: false
2. App launches with existing login state and user data preserved
3. Run test scenarios
4. Close session (app remains installed with data intact)

Testing an iOS app on real device:

Copy code
Test my iOS app on my physical iPhone:
1. Start app session with:
   - platform: iOS
   - appPath: /path/to/MyApp.ipa
   - deviceName: My iPhone
   - udid: 00008030-001234567890ABCD (your device's UDID)
   - platformVersion: 17.0
2. Run your test scenario
3. Close the session

Testing an Android app:

Copy code
Test my Android app /path/to/app.apk on the Pixel_6_API_34 emulator:
1. Start the app with auto-grant permissions
2. Get visible elements (use inViewportOnly: false to see all elements)
3. Swipe up to scroll
4. Tap on the "Settings" button using text matching
5. Verify the settings screen is displayed

Advanced element detection:

Copy code
Test my app and debug layout issues:
1. Start the app session
2. Get visible elements with includeContainers: true to see the layout hierarchy
3. Analyze ViewGroup, FrameLayout, and ScrollView containers
4. Use inViewportOnly: false to find off-screen elements that need scrolling

Hybrid app testing (switching contexts):

Copy code
Test my hybrid app:
1. Start the Android app session
2. Tap "Open Web" button in native context
3. List available contexts
4. Switch to WEBVIEW context
5. Click the login button using CSS selector
6. Switch back to NATIVE_APP context
7. Verify we're back on the home screen

Important Notes

⚠️ Session Management:

• Only one session (browser OR app) can be active at a time
• Always close sessions when done to free system resources
• To switch between browser and mobile, close the current session first
• Use close_session({ detach: true }) to disconnect without terminating the session on the Appium server
• State preservation can be controlled with noReset and fullReset parameters during session creation
• Sessions created with noReset: true or without appPath will automatically detach on close
• Sessions adopted with attach_session always detach on close unless detach: false is explicitly requested

⚠️ Task Planning:

• Break complex automation into smaller, focused operations
• Claude may consume message limits quickly with extensive automation

⚠️ Mobile Automation:

• Appium server must be running before starting mobile sessions
• Ensure emulators/simulators are running and devices are connected
• iOS automation requires macOS with Xcode installed
• iOS Real Devices: Testing on physical iOS devices requires the device's UDID (40-character unique identifier). See Prerequisites section for how to find your UDID

Selector Syntax Quick Reference

Web (CSS/XPath):

• CSS: button.my-class, #element-id
• XPath: //button[@class='my-class']
• Text: button=Exact text, a*=Contains text

Mobile (Cross-Platform):

• Accessibility ID: ~loginButton (works on both iOS and Android)
• Android UiAutomator: android=new UiSelector().text("Login")
• iOS Predicate: -ios predicate string:label == "Login" AND visible == 1
• XPath: //android.widget.Button[@text="Login"]

Advanced Features

App State Preservation

State Preservation with noReset/fullReset: Control app state when creating new sessions using the noReset and fullReset parameters:

Example with state preservation:

javascriptCopy code
// Preserve login state between test runs
start_session({
    platform: 'android',
    appPath: '/path/to/app.apk',
    deviceName: 'emulator-5554',
    noReset: true,         // Don't reset app state
    fullReset: false,      // Don't uninstall
    autoGrantPermissions: true,
    capabilities: {
        'appium:chromedriverExecutable': '/path/to/chromedriver',
        'appium:autoWebview': true
    }
})
// App launches with existing user data, login tokens, preferences intact

Detach from Sessions: The close_session tool supports a detach parameter that disconnects from the session without terminating it on the Appium server:

javascriptCopy code
// Detach without killing the session
close_session({detach: true})

// Explicit session termination (closes the app and removes session)
close_session({detach: false})

Sessions created with noReset: true or without appPath will automatically detach on close. Sessions adopted with attach_session are externally managed and also detach by default; pass detach: false only when the MCP should deliberately terminate the existing remote session.

This is particularly useful when:

• Preserving app state for manual testing continuation
• Debugging multi-step workflows (leave session running between tool invocations)
• Testing scenarios where you want the app to remain installed and in current state

Smart Element Detection

• Platform-specific element classification: Automatically identifies interactable elements vs layout containers
  • Android: Button, EditText, CheckBox vs ViewGroup, FrameLayout, ScrollView
  • iOS: Button, TextField, Switch vs View, StackView, CollectionView
• Multiple locator strategies: Each element provides accessibility ID, resource ID, text, XPath, and platform-specific selectors
• Viewport filtering: Control whether to get only visible elements or all elements including off-screen
• Layout debugging: Optionally include container elements to understand UI hierarchy

Automatic Permission & Alert Handling

Both iOS and Android sessions now support automatic handling of system permissions and alerts:

• autoGrantPermissions (default: true): Automatically grants app permissions (camera, location, etc.)
• autoAcceptAlerts (default: true): Automatically accepts system alerts and dialogs
• autoDismissAlerts (optional): Set to true to dismiss alerts instead of accepting them

This eliminates the need to manually handle permission popups during automated testing.

Technical Details

• Built with: TypeScript, WebDriverIO, Appium
• Browser Support: Chrome, Firefox, Edge (headed/headless, automated driver management), Safari (headed only; macOS)
• Mobile Support: iOS (XCUITest) and Android (UiAutomator2/Espresso)
• Protocol: Model Context Protocol (MCP) for Claude Desktop integration
• Session Model: Single active session (browser or mobile app)
• Data Format: TOON (Token-Oriented Object Notation) for efficient LLM communication
• Element Detection: XML-based page source parsing with intelligent filtering and multi-strategy locator generation

Session Recording & Code Export

Every tool call is automatically recorded to a session history. You can inspect sessions and export runnable code via MCP resources — no extra tool calls needed:

• wdio://sessions — lists all recorded sessions with type, timestamps, and step count
• wdio://session/current/steps — step log for the active session
• wdio://session/current/code — generated runnable WebdriverIO JS for the active session
• wdio://session/{sessionId}/steps — step log for any past session by ID
• wdio://session/{sessionId}/code — generated JS for any past session by ID

The generated script reconstructs the full session — including capabilities, navigation, clicks, and inputs — as a standalone import { remote } from 'webdriverio' file. For cloud provider sessions it includes the full try/catch/finally with automatic session result marking via the provider's REST API.

Trace Recording

Passing trace: true to start_session produces a Playwright-compatible .trace zip in the .trace/ directory when the session closes. The zip is playable at player.vibium.dev and shows a filmstrip of screenshots alongside the action timeline.

How screenshots are timed

Appium's takeScreenshot round-trip takes 700–1300 ms on a local emulator, which is long enough for the previous action's animations to settle. We exploit this: each screenshot is captured before the next action fires, so what the Appium server returns is already the settled result of the prior action.

The tricky part is making the trace player show that screenshot under the right action. The player associates a screencast-frame event with whichever action's time window contains the frame's timestamp field. If the timestamp is set to "now" (capture time), it falls before the current action's startTime and the player labels it as the before state of the next action — one action out of sync.

The fix: stamp each screencast-frame with lastAfterEndTime — the endTime of the action that just completed. That places the frame inside the previous action's window, so the player shows it as the result of that action, not the precursor to the next one.

Copy code
Timeline (monotonic ms):

  prev.endTime ← frame timestamp stamped here
        │
        │   [screenshot captured here — shows settled state after prev action]
        │
  curr.startTim