TestingBot MCP Server
Official# TestingBot MCP Server
[](https://www.npmjs.com/package/@testingbot/mcp-server)
[](https://opensource.org/licenses/MIT)
[](https://lobehub.com/mcp/testingbot-mcp-server)
TestingBot's official Model Context Protocol (MCP) server implementation. This server enables AI assistants to interact with TestingBot's testing infrastructure, allowing you to manage tests, browsers, devices, and more through conversational interfaces.
## โก One-Click Install
[](https://testingbot.com/mcp/install?client=vscode) [](https://testingbot.com/mcp/install?client=cursor)
**Claude Desktop**: download `testingbot-mcp-server-<version>.mcpb` from the [latest release](https://github.com/testingbot/mcp-server/releases/latest) and double-click it. Claude will set up the TestingBot MCP server automatically.
After install, set your credentials from [TestingBot account settings](https://testingbot.com/members/user/security).
## Features
- ๐ฑ **App Automate** - Run Maestro, Espresso, and XCUITest suites on real devices and emulators, straight from the agent
- ๐ฅ๏ธ **Live Testing** - Start interactive manual testing sessions on real browsers and devices
- ๐ **Browser & Device Management** - Query available browsers and real devices
- ๐งช **Test Management** - Create, retrieve, update, and delete tests with comprehensive details
- ๐ฆ **Storage Management** - Upload and manage mobile app files (APK, IPA, ZIP)
- ๐ธ **Screenshot Testing** - Take cross-browser screenshots
- ๐๏ธ **Build Management** - Organize tests into builds
- ๐ค **User Account** - View and update account information
- ๐ฅ **Team Management** - Manage team settings and team members
- ๐ **Chrome DevTools Protocol** - Create CDP sessions for advanced browser automation
- ๐ **Tunnel Management** - Start, stop and manage TestingBot tunnels to test localhost and private sites
## Prerequisites
- **Node.js** >= 20 (recommended: 22.15.0 or later)
- **TestingBot Account** with API credentials
- An MCP-compatible client (Claude Desktop, VS Code with Continue, Cursor, etc.)
## Installation
### Quick Setup
Install the MCP server globally:
```bash
npm install -g @testingbot/mcp-server
```
Or add to your project:
```bash
npm install @testingbot/mcp-server
```
### Configuration
#### Recommended: log in with `tb_login` (no key/secret to copy)
You don't have to find and paste API credentials. Install the server, leave the
credentials blank, and just ask the agent:
> Log me in to TestingBot.
The agent calls the **`tb_login`** tool. It uses OAuth-style browser
authentication โ no secret ever leaves the browser โ and picks the right
mechanism for your environment automatically:
- **Browser (loopback, the default on a desktop):** your browser opens to a
TestingBot authorization page. Sign in if needed and click **Authorize**; the
credentials are captured on a local callback and the agent finishes the
sign-in. Nothing to type. (RFC 8252.)
- **Device code (automatic fallback on SSH / remote / dev containers / web):**
the tool prints a short URL and code instead:
1. Open the URL in your browser (e.g. `https://testingbot.com/device`).
2. Enter the code, sign in if needed, and click **Authorize**.
3. Tell the agent you've authorized โ it calls `tb_login` again to finish.
(RFC 8628.)
You can force a mechanism by passing `mode: "loopback"` or `mode: "device"` to
`tb_login` (the default is `"auto"`).
Your credentials are written to `~/.testingbot/credentials` (mode `0600`) and
used by every subsequent tool call โ no restart and no JSON editing. This works
in any MCP client (VS Code, Cursor, Cline, Claude Desktop, โฆ). To use a different
account, set `TESTINGBOT_PROFILE`; to relocate the file, set
`TESTINGBOT_CONFIG_DIR`.
#### Environment Variables
Alternatively (or for CI, where env vars take precedence over the file), set your
TestingBot credentials as environment variables:
```bash
export TESTINGBOT_KEY="your-api-key"
export TESTINGBOT_SECRET="your-api-secret"
```
Or create a `.env` file in your project:
```
TESTINGBOT_KEY=your-api-key
TESTINGBOT_SECRET=your-api-secret
```
The key also accepts the aliases `TB_KEY` / `TESTINGBOT_USERNAME`, and the secret
accepts `TB_SECRET` / `TESTINGBOT_ACCESS_KEY`.
#### How credentials are resolved
Credentials are resolved in this order, highest priority first:
1. **Environment variables** (above) โ both a key and a secret must be set. This
lets CI and per-client config override everything else.
2. **The `~/.testingbot/credentials` file** written by `tb_login`, for the active
profile (`default`, or `TESTINGBOT_PROFILE`).
3. **Neither set:** the server starts in a degraded mode where every tool except
`tb_login` reports "Run tb_login to authenticate" โ so first run is
self-healing rather than a hard error.
#### Claude Desktop
Add to `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%/Claude/claude_desktop_config.json` (Windows):
```json
{
"mcpServers": {
"testingbot": {
"command": "npx",
"args": ["-y", "@testingbot/mcp-server"],
"env": {
"TESTINGBOT_KEY": "your-api-key",
"TESTINGBOT_SECRET": "your-api-secret"
}
}
}
}
```
#### VS Code (with Continue extension)
Add to `.continue/config.json`:
```json
{
"mcpServers": [
{
"name": "testingbot",
"command": "npx",
"args": ["-y", "@testingbot/mcp-server"],
"env": {
"TESTINGBOT_KEY": "your-api-key",
"TESTINGBOT_SECRET": "your-api-secret"
}
}
]
}
```
#### Cursor
Add to `.cursor/mcp.json`:
```json
{
"mcpServers": {
"testingbot": {
"command": "npx",
"args": ["-y", "@testingbot/mcp-server"],
"env": {
"TESTINGBOT_KEY": "your-api-key",
"TESTINGBOT_SECRET": "your-api-secret"
}
}
}
}
```
## ๐ค Let an AI Agent Drive a Remote Browser or Device
This server already includes the automation tools from [`@testingbot/automation-mcp`](https://github.com/testingbot/automation-mcp) โ installing this package gives the agent the ability to **actually drive** a remote browser or real mobile device on TestingBot's grid, not just manage resources.
Just ask:
> Open Safari 17 on macOS on TestingBot. Navigate to example.com and click the "More information" link.
The agent will call `tb_openBrowser`, `tb_navigate`, `tb_snapshot`, `tb_click` โ and surface a **live-view URL** so you can watch it in real time. For real iOS / Android devices, the bundled [`appium-mcp`](https://github.com/appium/appium-mcp) child gives the agent the full `appium_*` toolbelt (`appium_session_management`, `appium_gesture`, `appium_set_value`, `appium_screenshot`, ~30 more) โ no separate install or config required. Just ask: _"Open an iPhone 15 Pro on iOS 17 on TestingBot, tap the login button, take a screenshot."_ See the [`@testingbot/automation-mcp` README](https://github.com/testingbot/automation-mcp#readme) for the full tool list, caveats (sessions are metered), and configuration options.
## ๐ Testing localhost and private sites
TestingBot's browsers and devices run in the cloud, so they can't open `localhost`, private IPs (`10.x`, `192.168.x`, โฆ) or intranet hosts directly. To reach them, they need a [TestingBot Tunnel](https://testingbot.com/support/tunnel) running on your machine. Just ask:
> Start a tunnel and open http://localhost:3000 in Chrome on Windows 11.
The agent calls `startTunnel`, which needs **Java 17+**. The tool returns the tunnel's identifier and ID, and the agent passes them on:
| Run | How it uses the tunnel |
| --- | --- |
| `tb_openBrowser` (agent-driven browser) | `tunnelIdentifier` (**required**: sessions without it go direct). Or set `TESTINGBOT_TUNNEL_IDENTIFIER` |
| `appium_session_management` (native apps) | `"tb:options": {"tunnel-identifier": "โฆ"}` in the capabilities |
| `startLiveSession` / `startDesktopLiveSession` / `startMobileLiveSession` | `tunnelId` (the numeric ID) |
| `runMaestroTest` | `tunnelIdentifier`. Without it, the account's newest running tunnel is used |
| `takeScreenshot`, `runAppAutomateTest` (Espresso / XCUITest) | The account's newest running tunnel is used automatically |
**Android apps and flows:** Android never sends `localhost` / `127.0.0.1` through a proxy, so on an Android device or emulator those URLs reach the device itself, not your machine. Use a hostname that resolves to 127.0.0.1 on your side instead, such as `http://myapp.localtest.me:3000`. The tunnel resolves it on your machine.
`stopTunnel` stops the tunnel. Tunnels started by the server are also stopped when the server exits. If you'd rather run the tunnel yourself, start it with `--tunnel-identifier <name>` and give the agent that name.
## Available Tools
### Live Testing
#### `startLiveSession`
Start an interactive live testing session on TestingBot. Returns a link that opens the session in your TestingBot dashboard, where you confirm it with Start. Mobile sessions run on emulators/simulators; real-device live sessions are started from the dashboard.
**Parameters:**
- `platformType`: "desktop" or "mobile"
- `desiredURL`: The URL to open
- `desiredOS`: Operating system (Windows, Mac, Linux for desktop; android, ios for mobile)
- `desiredOSVersion`: OS version: Windows `11`/`10`/`8.1`/`8`; macOS by name (`Sonoma`) or number (`14`); or a platform code like `WIN11`
- `desiredBrowser` (desktop only): Browser name (chrome, firefox, safari, edge, ie)
- `desiredBrowserVersion` (desktop only): Browser version or "latest"
- `desiredDevice` (mobile only): Device name (e.g., "iPhone 14", "Galaxy S23")
- `tunnelId` (optional): Numeric ID of a running tunnel, needed for localhost / private URLs (see [Testing localhost and private sites](#-testing-localhost-and-private-sites))
**Example prompts:**
> "Start a live testing session on Chrome latest with Windows 11 for https://example.com"
> "Open https://myapp.com on iPhone 14 with iOS 16 for manual testing"
#### `startDesktopLiveSession`
Convenience tool to start a desktop browser live testing session.
**Parameters:**
- `desiredURL`: The URL to open
- `desiredOS`: Windows, Mac, or Linux
- `desiredOSVersion`: OS version
- `desiredBrowser`: chrome, firefox, safari, edge, or ie
- `desiredBrowserVersion` (optional): Browser version or "latest" (default: latest)
- `tunnelId` (optional): Numeric ID of a running tunnel, for localhost / private URLs
**Example prompt:**
> "Start desktop live session on Firefox 120 with Mac Monterey for https://example.com"
#### `startMobileLiveSession`
Convenience tool to start a mobile device live testing session.
**Parameters:**
- `desiredURL`: The URL to open
- `desiredOS`: android or ios
- `desiredOSVersion`: OS version
- `desiredDevice`: Device name
- `tunnelId` (optional): Numeric ID of a running tunnel, for localhost / private URLs
**Example prompt:**
> "Start mobile live session on Galaxy S23 with Android 13 for https://example.com"
### Browser & Device Management
#### `getBrowsers`
Get list of available browsers and platforms for testing.
**Parameters:**
- `type` (optional): Filter by "web" or "mobile"
**Example prompt:**
> "Show me all available browsers for testing"
> "What mobile browsers are available?"
#### `getDevices`
Get list of available mobile devices (real devices and simulators).
**Example prompt:**
> "List all available iOS devices"
> "Show me Android devices for testing"
### Test Management
#### `getTests`
Retrieve a list of recent tests with pagination.
**Parameters:**
- `offset` (optional): Pagination offset (default: 0)
- `limit` (optional): Number of tests to retrieve (default: 10, max: 100)
**Example prompt:**
> "Show me my last 20 tests"
> "Get recent test results"
#### `getTestDetails`
Get comprehensive details about a specific test including:
- Test status, browser, platform, and timing information
- Video recording URL and screenshot URLs
- Selenium, browser, and Appium logs
- Test execution steps with commands and timestamps
- Network logs, exception logs, and JS errors
**Parameters:**
- `sessionId`: The session ID of the test
**Example prompt:**
> "Get details for test session abc123"
> "Show me the video and logs for test xyz789"
> "Show execution steps for test abc123"
#### `updateTest`
Update test metadata such as name, status, or build.
**Parameters:**
- `sessionId`: The session ID of the test
- `name` (optional): New name for the test
- `status` (optional): "passed" or "failed"
- `build` (optional): Build identifier
- `extra` (optional): Additional metadata (JSON string)
**Example prompt:**
> "Mark test abc123 as passed"
> "Update test xyz789 with name 'Login Flow Test'"
#### `deleteTest`
Delete a test by session ID.
**Parameters:**
- `sessionId`: The session ID of the test to delete
**Example prompt:**
> "Delete test abc123"
#### `stopTest`
Stop a running test by session ID.
**Parameters:**
- `sessionId`: The session ID of the test to stop
**Example prompt:**
> "Stop test abc123"
### AI Failure Analysis
Explanations from [TestingBot AI Insights](https://testingbot.com/support/ai/insights): summary, root cause, timeline, log evidence, suggested fixes, a failure category, the likely owner and a confidence score. The account owner must first enable **AI Analysis** in Account Settings, which records consent to sending curated, secret-masked test logs to TestingBot's AI provider; until then both tools say how to turn it on and nothing is sent.
#### `analyzeTestFailure`
Start (or reuse) an AI failure analysis of a finished test and wait for the result. A current analysis is returned without extra cost.
**Parameters:**
- `sessionId`: Session ID (or numeric test ID)
- `force` (optional): Run a fresh analysis even if one exists
- `waitSeconds` (optional): How long to wait for the result (default 90, 0 returns immediately)
**Example prompt:**
> "Why did test abc123 fail?"
#### `getAiFailureAnalysis`
Read an existing analysis. Never starts one.
**Parameters:**
- `sessionId`: Session ID (or numeric test ID)
### Build Management
#### `getBuilds`
Get a list of builds with pagination.
**Parameters:**
- `offset` (optional): Pagination offset (default: 0)
- `limit` (optional): Number of builds to retrieve (default: 10, max: 100)
**Example prompt:**
> "Show me my recent builds"
#### `getTestsForBuild`
Get all tests associated with a specific build ID.
**Parameters:**
- `buildId`: The build ID
**Example prompt:**
> "Show all tests for build 12345"
#### `deleteBuild`
Delete a build and all its associated tests.
**Parameters:**
- `buildId`: The build ID to delete
**Example prompt:**
> "Delete build 12345"
### Storage Management
#### `uploadFile`
Upload a local file (APK, IPA, or ZIP) to TestingBot storage.
**Parameters:**
- `localFilePath`: Path to the file to upload
**Example prompt:**
> "Upload /path/to/app.apk to TestingBot"
#### `uploadRemoteFile`
Upload a file from a remote URL to TestingBot storage.
**Parameters:**
- `remoteUrl`: URL of the file to upload
**Example prompt:**
> "Upload https://example.com/app.ipa to TestingBot"
#### `getStorageFiles`
List all files in TestingBot storage.
**Parameters:**
- `offset` (optional): Pagination offset (default: 0)
- `limit` (optional): Number of files to retrieve (default: 10, max: 100)
**Example prompt:**
> "Show me all uploaded apps"
#### `deleteStorageFile`
Delete a file from TestingBot storage.
**Parameters:**
- `appUrl`: The app_url of the file to delete
**Example prompt:**
> "Delete app tb://app123 from storage"
### App Automate โ Maestro
Run [Maestro](https://maestro.dev) flows on TestingBot real devices, emulators, and simulators. Runs are **asynchronous**: start one, poll its status, then fetch results โ the agent is never blocked while devices execute.
#### `uploadMaestroApp`
Upload the app under test (`.apk`, `.aab`, `.ipa`, or zipped `.app`). Uploads are deduplicated by checksum โ re-running with the same binary is instant.
**Parameters:**
- `localFilePath`: Local path to the app file
Returns a **Project ID** used by the other Maestro tools.
**Example prompt:**
> "Upload ./build/app.apk for Maestro testing"
#### `uploadMaestroFlows`
Upload the Maestro flow YAML files for a project. Accepts a directory (structure preserved, so `runFlow` subflows keep working), a pre-built `.zip`, or **inline YAML** written by the agent. Replaces previously uploaded flows.
**Parameters:**
- `projectId`: Project ID from `uploadMaestroApp`
- `flowsPath` (optional): Directory or `.zip` of flow YAML files
- `flows` (optional): Inline flows as `[{ fileName, content }]`
#### `runMaestroTest`
Start a Maestro run. Returns immediately with the run ID โ poll with `getMaestroRunStatus`.
**Parameters:**
- `projectId`: Project ID
- `platformName`: "Android" or "iOS"
- `deviceName` (optional): Device name, wildcards/regex supported (e.g. `"Pixel 8"`, `"iPhone 1[4-6]"`, `"*"`)
- `version` (optional): OS version
- `realDevice` (optional): Run on a physical device (required for `.ipa`)
- `name` (optional): Run name shown in the dashboard
- `env` (optional): Environment variables passed to the flows
- `includeTags` / `excludeTags` (optional): Filter flows by Maestro tags
- `shardSplit` (optional): Split flows across parallel devices
- `otherApps` (optional): Up to 4 companion app URLs to install alongside the app
- `tunnelIdentifier` (optional): Run through this tunnel, so the app can reach localhost / private backends
**Example prompt:**
> "Run my Maestro flows on a real Pixel 8 with Android 14"
#### `getMaestroRunStatus`
Poll run progress. Lists per-flow statuses (WAITING, READY, DONE, FAILED, CANCELLED), failures first.
#### `getMaestroRunResults`
Final results of a completed run: per-flow outcomes plus the JUnit XML report.
#### `getMaestroFlowDetails`
Deep-dive into one flow: error messages, step-level JUnit report, and links to the video, screenshots, and device logs. This is the debugging tool โ fetch it for a failed flow, fix the YAML, then retry.
**Parameters:** `projectId`, `runId`, `flowId`
#### `cancelMaestroRun` / `retryMaestroRun`
Cancel a running run, or retry a run โ pass `flowId` to retry only a single failed flow.
#### `uploadMaestroCompanionApp`
Upload a supplementary app installed alongside the app under test. Returns a `tb://` URL for `runMaestroTest`'s `otherApps`.
#### `listMaestroProjects` / `listMaestroRuns`
Discover existing projects (app, flows, run IDs) and runs across the account. `listMaestroRuns` accepts a `buildName` to look up runs by name โ e.g. "rerun whatever failed in last night's `nightly` build".
#### `maestroCheatSheet`
Returns a Maestro YAML syntax reference (commands, selectors, flow structure, and the TestingBot workflow). Agents call this before writing or fixing flows.
**Example workflow:**
```
User: "Run the Maestro flows in ./flows against ./app.apk on a real Pixel 8. Fix any failures."
Assistant: [uploadMaestroApp โ uploadMaestroFlows โ runMaestroTest]
Assistant: [polls getMaestroRunStatus until complete]
Assistant: [getMaestroFlowDetails for the failed flow, reads the error + screenshots]
Assistant: [fixes the flow YAML, uploadMaestroFlows, retryMaestroRun]
```
### App Automate โ Espresso & XCUITest
Run pre-built Espresso (Android) and XCUITest (iOS) suites on TestingBot devices. Same async pattern: start, poll, fetch results.
#### `uploadAppAutomateApp`
Upload the app under test: `.apk`/`.aab` for Espresso, `.ipa` for XCUITest.
**Parameters:** `framework` ("espresso" or "xcuitest"), `localFilePath`
#### `uploadAppAutomateTests`
Upload the test suite: the instrumented test `.apk` for Espresso, or the zipped XCUITest runner bundle.
**Parameters:** `framework`, `projectId`, `localFilePath`
#### `runAppAutomateTest`
Start a run. Returns immediately with the run ID.
**Parameters:**
- `framework`, `projectId`
- `deviceName`, `version`, `realDevice`, `name`, `build` (optional): device selection and metadata
- `testClasses`, `skipTestClasses`, `packages`, `annotations`, `testRunner` (optional, Espresso only): test filtering
- `locale`, `timeZone` (optional)
**Example prompt:**
> "Run the Espresso suite on a real Galaxy S23, only the com.example.LoginTest class"
#### `getAppAutomateRunStatus` / `getAppAutomateRunResults`
Poll run progress, then fetch per-run outcomes and the JUnit XML report. Each run's Appium session ID is surfaced, so `getTestDetails` / `getFailureLogs` work for logs and video.
### Visual Testing Review
Review [hosted visual tests](https://testingbot.com/support/visual-testing/ci#review-api): find failed runs, see what changed, approve or reject. When the account owner has turned on AI region names, each changed region carries a one-line AI description of the change; reading a run never sends anything to an AI provider.
#### `getVisualTests`
List hosted visual tests. **Parameters:** `offset`, `limit` (optional)
#### `getVisualRuns`
List visual runs, newest first. **Parameters:** `testId`, `state` (`failed`, `unreviewed`, `passed`), `offset`, `limit` (all optional)
#### `explainVisualRun`
What changed in a run: regions with names, AI change descriptions, text and CSS changes, and image links. **Parameters:** `runId`
**Example prompt:**
> "Which visual tests failed today, and what changed on the pricing page?"
#### `reviewVisualRun`
Approve or reject a run. **Parameters:** `runId`, `decision` (`approve`/`reject`), `makeBaseline` (optional, with approve)
### Accessibility Testing
Results of scheduled [accessibility scans](https://testingbot.com/support/accessibility), scored like the dashboard: WCAG A/AA violations decide pass/fail; needs-review items and best-practice recommendations do not. Automated checks cover only part of WCAG.
#### `getAccessibilityTests`
Scheduled accessibility tests with their latest result. **Parameters:** `offset`, `limit` (optional)
#### `getAccessibilityRuns`
Scans of one test, newest first. **Parameters:** `testId`, `offset`, `limit`
#### `getAccessibilityFindings`
Failing rules with the elements (selector, HTML, how to fix) and what is new or resolved since the previous scan. **Parameters:** `runId`, `elementsPerRule` (optional, default 5)
**Example prompt:**
> "What accessibility issues does our checkout page have, and which are new since the last scan?"
#### `runAccessibilityTest`
Start a scan now. **Parameters:** `testId`
### Screenshot Testing
#### `takeScreenshot`
Take screenshots of a URL across multiple browsers and platforms.
**Parameters:**
- `url`: The URL to screenshot
- `browsers`: Array of browser configurations
- `browserName`: Browser name (chrome, firefox, safari, etc.)
- `version` (optional): Browser version or "latest"
- `os`: Operating system (WIN11, MAC, etc.)
- `resolution` (optional): Screen resolution (default: "1920x1080")
- `waitTime` (optional): Seconds to wait before screenshot (default: 5, max: 60)
- `fullPage` (optional): Capture full page or viewport (default: false)
For a localhost / private URL, the screenshots go through the account's newest running tunnel. If no tunnel is running, the tool says so instead of taking the screenshots.
**Example prompt:**
> "Take a screenshot of https://example.com on Chrome and Firefox"
> "Screenshot my homepage on mobile devices"
#### `retrieveScreenshots`
Retrieve screenshot results by screenshot ID.
**Parameters:**
- `screenshotId`: The screenshot ID from takeScreenshot
**Example prompt:**
> "Get screenshots for job abc123"
#### `getScreenshotList`
Get a list of all screenshot jobs.
**Parameters:**
- `offset` (optional): Pagination offset (default: 0)
- `limit` (optional): Number of jobs to retrieve (default: 10, max: 100)
**Example prompt:**
> "Show me my recent screenshot jobs"
### User Management
#### `getUserInfo`
Get current user account information including minutes used, plan details, and limits.
**Example prompt:**
> "Show my account information"
> "How many minutes have I used?"
#### `updateUserInfo`
Update user account information.
**Parameters:**
- `firstName` (optional): First name
- `lastName` (optional): Last name
- `email` (optional): Email address
**Example prompt:**
> "Update my email to newemail@example.com"
### Team Management
#### `getTeam`
Get team information including concurrency: parallel sessions allowed, running right now and queued (waiting for a free slot), for VMs and physical devices. Counts are team-wide and match the dashboard.
**Example prompt:**
> "Show my team information"
> "What are my team's concurrency limits?"
> "How many of my tests are running or queued right now?"
#### `getUsersInTeam`
List all users in your team with their roles and access levels.
**Parameters:**
- `offset` (optional): Pagination offset (default: 0)
- `limit` (optional): Number of users to retrieve (default: 10, max: 100)
**Example prompt:**
> "Show me all team members"
> "List users in my team"
#### `getUserFromTeam`
Get detailed information about a specific team member by user ID.
**Parameters:**
- `userId`: The user ID to retrieve
**Example prompt:**
> "Show details for user 12345"
### Chrome DevTools Protocol (CDP)
#### `createCdpSession`
Create a Chrome DevTools Protocol session for advanced browser automation. Returns a WebSocket URL to connect to the browser.
**Parameters:**
- `browserName`: Browser name (chrome, firefox, safari, edge, etc.)
- `browserVersion` (optional): Browser version or "latest" (default: latest)
- `platform`: Operating system platform
- `screenResolution` (optional): Screen resolution (e.g., "1920x1080")
- `name` (optional): Session name
- `build` (optional): Build identifier
**Example prompt:**
> "Create a CDP session on Chrome latest with Windows 11"
> "Start a CDP session on Firefox 120 for automation"
### Tunnel Management
TestingBot's browsers and devices run in the cloud. To test `localhost`, private IPs or intranet sites, they need a [TestingBot Tunnel](https://testingbot.com/support/tunnel) running on your machine. These tools can start one for you.
#### `startTunnel`
Starts a TestingBot Tunnel on the machine running the MCP server and waits until it is ready. Returns the tunnel's identifier and numeric ID. The tunnel stops with `stopTunnel` or when the MCP server exits. On first use the tunnel is downloaded, and it needs **Java 17 or newer**.
**Parameters:**
- `tunnelIdentifier` (optional): Name for the tunnel. If you don't give one, a unique one is generated.
- `shared` (optional): Share the tunnel with your team
- `noBump` (optional): Do not re-sign HTTPS traffic (for certificate pinning or client certificates)
- `noCache` (optional): Disable caching in the tunnel
**Example prompt:**
> "Start a tunnel and open http://localhost:3000 in Chrome on Windows 11"
#### `stopTunnel`
Stops a tunnel started with `startTunnel`.
**Parameters:**
- `tunnelIdentifier` (optional): The tunnel to stop. Can be left out when only one tunnel is running.
- `all` (optional): Stop every tunnel this server started
#### `getTunnelList`
Get a list of all active TestingBot tunnels on your account, with their identifiers. Tunnels allow testing websites behind firewalls or on your local machine.
**Example prompt:**
> "Show me all active tunnels"
> "List my TestingBot tunnels"
#### `deleteTunnel`
Delete an active TestingBot tunnel by ID. This terminates the tunnel connection.
**Parameters:**
- `tunnelId`: The tunnel ID to delete
**Example prompt:**
> "Delete tunnel 12345"
## Prompts, Resources and Agent Skills
Besides tools, the server offers MCP **prompts**: task recipes your client can show as slash commands:
| Prompt | Arguments | What it does |
| --- | --- | --- |
| `why-did-my-test-fail` | `sessionId` | Test status, AI failure analysis when enabled, failing log lines, likely cause and fixes |
| `summarise-build` | `buildId` | Pass/fail counts, failures grouped by shared cause, what to fix first |
| `find-flaky-tests` | `timeRange` (optional, default `30d`) | Tests that both passed and failed, and why each is unstable |
and read-only **resources** a client can attach to a conversation:
| URI | Content |
| --- | --- |
| `testingbot://tests/{sessionId}` | Test details (JSON) |
| `testingbot://tests/{sessionId}/logs/{logType}` | One log file, last 200 KB (`selenium`, `browser`, `chrome`, `vm`, `appium`) |
| `testingbot://builds/{buildId}` | Tests in a build (JSON) |
| `testingbot://insights/flaky-tests` | Flaky tests over the last 30 days; `?time_range=7d` to change it |
The same recipes ship as agent skills in [`skills/`](skills/) (one `SKILL.md` per task) for agents that load skills from files.
## Usage Examples
### Example 1: Running Cross-Browser Tests
```
User: "Show me all available browsers for web testing"
Assistant: [Lists all web browsers]
User: "Take a screenshot of https://myapp.com on Chrome latest and Firefox latest on Windows 11"
Assistant: [Creates screenshot job and returns ID]
User: "Get the screenshots for job abc123"
Assistant: [Returns screenshot URLs]
```
### Example 2: Managing Mobile App Tests
```
User: "Upload my app from https://example.com/app.apk"
Assistant: [Uploads app and returns tb://app123]
User: "Show me all my uploaded apps"
Assistant: [Lists storage files]
User: "Show me the last 10 tests"
Assistant: [Shows recent test results]
```
### Example 3: Agent-Driven Maestro Testing on Real Devices
```
User: "Write a Maestro flow that logs into my app and run it on a real Pixel 8"
Assistant: [maestroCheatSheet for syntax, writes the flow YAML]
Assistant: [uploadMaestroApp ./app.apk โ project 42]
Assistant: [uploadMaestroFlows with the inline flow]
Assistant: [runMaestroTest on Pixel 8 with realDevice: true โ run 101]
Assistant: [polls getMaestroRunStatus, then getMaestroRunResults]
User: "The checkout flow failed โ figure out why and fix it"
Assistant: [getMaestroFlowDetails โ reads error + screenshots + video]
Assistant: [fixes the YAML, uploadMaestroFlows, retryMaestroRun]
```
### Example 4: Build Management
```
User: "Show me my recent builds"
Assistant: [Lists recent builds]
User: "Show all tests for build 12345"
Assistant: [Lists tests in that build]
User: "Mark test xyz789 as passed with build name 'Release 1.0'"
Assistant: [Updates test]
```
### Example 5: Team & Tunnel Management
```
User: "Show me my team information"
Assistant: [Shows team concurrency limits and settings]
User: "List all team members"
Assistant: [Lists all users in the team]
User: "Show me all active tunnels"
Assistant: [Lists active TestingBot tunnels]
User: "Start a tunnel and check that http://localhost:3000 loads in Safari on macOS"
Assistant: [Starts a tunnel, opens Safari with that tunnel, loads the page]
User: "Stop the tunnel"
Assistant: [Stops the tunnel it started]
```
### Example 6: CDP Automation
```
User: "Create a CDP session on Chrome latest with Windows 11"
Assistant: [Creates session and returns WebSocket URL]
User: "Show me test execution steps for session abc123"
Assistant: [Shows detailed command history with timestamps]
```
## Development
### Setup
```bash
# Clone the repository
git clone https://github.com/testingbot/mcp-server.git
cd testingbot-mcp-server
# Install dependencies
npm install
# Set up environment variables
cp .env.example .env
# Edit .env with your credentials
```
### Commands
```bash
# Build the project (runs lint, format, test, and compile)
npm run build
# Development mode with watch
npm run dev
# Run tests
npm test
# Lint code
npm run lint
# Format code
npm run format
# Build the .mcpb bundle for Claude Desktop (see section below)
npm run mcpb
```
### Running Locally
```bash
# Development mode
npm run dev
# Production mode
npm start
```
### Testing with MCP Inspector
Use the MCP Inspector to test tools during development:
```bash
npx @modelcontextprotocol/inspector node dist/index.js
```
Then open http://localhost:5173 in your browser.
### Building the `.mcpb` bundle (Claude Desktop release)
The `.mcpb` archive is what users double-click to install in Claude Desktop. The installer reads `manifest.json#user_config` and shows a credential prompt โ no JSON editing or env-var fiddling.
```bash
# Prerequisites: mcpb CLI (one-time)
npm install -g @anthropic-ai/mcpb
# Build the bundle
npm run mcpb
```
This runs `scripts/build-mcpb.sh`, which:
1. Runs the full quality pipeline (`npm run build` โ lint โ format โ test โ tsc).
2. Stages a clean tree in a temp directory (only `dist/`, `manifest.json`, `package.json`, `LICENSE`, `README.md`, `icon.png` if present).
3. Runs `npm ci --omit=dev` inside staging โ dev deps (typescript, eslint, vitest) never enter the bundle. `@testingbot/automation-mcp` is pulled from npm at the version pinned in `package.json`.
4. Validates the manifest and runs `mcpb pack` honoring `.mcpbignore` (strips test fixtures, source maps, dependency changelogs).
5. Writes `releases/testingbot-mcp-server-<version>.mcpb` and runs `mcpb info` on it.
Output is roughly 160 MB โ most of that is the bundled `appium-mcp` mobile driver stack (XCUITest, UiAutomator2). The bundle is paid for once at install, not at runtime.
#### Sibling repo during development
For active development on both `@testingbot/mcp-server` and `@testingbot/automation-mcp` in lockstep, use `npm run dev:rebuild` from this repo. It builds both, then `npm link`s the sibling into this repo's `node_modules` so edits in `../testingbot-automation-mcp/src/` take effect on the next rebuild without a republish.
The link is one-shot โ once set up, `dev:rebuild` notices it's in place and just builds. If you ever run `npm install` in this repo, the link is replaced with the registry copy; rerun `dev:rebuild` to restore.
#### Releasing
The release flow has two coordinated publish steps:
1. **Publish `@testingbot/automation-mcp` to npm** (from `../testingbot-automation-mcp`):
```bash
npm version patch # or minor / major
npm publish # publishConfig.access is "public" โ handles scoped-pkg defaults
```
2. **Bump this repo to consume the new automation-mcp** and cut a release:
```bash
npm install @testingbot/automation-mcp@latest # updates package-lock.json
npm version patch # also syncs manifest.json + server.json
git push --follow-tags
# โ Create a GitHub release at the new tag. The Build MCPB workflow
# fires on release:published and attaches releases/*.mcpb to it.
```
The release CI (`.github/workflows/build-mcpb.yml`) runs `npm run mcpb` on `macos-latest` and uploads the bundle as both a workflow artifact (30-day retention) and a release asset. Manual re-build: trigger via the Actions tab โ "Build MCPB bundle" โ enter the tag.
macOS is chosen so the bundled `appium-xcuitest-driver` carries its native bits. If you only target Linux/Windows users, you can switch the runner to `ubuntu-latest` for ~10ร cheaper minutes โ at the cost of iOS support in the bundled mobile path. Browser tools (via WebDriver) work cross-platform regardless.
To validate the manifest without building:
```bash
npm run mcpb:validate
```
To inspect a built bundle:
```bash
mcpb info releases/testingbot-mcp-server-<version>.mcpb
mcpb unpack releases/testingbot-mcp-server-<version>.mcpb /tmp/unpacked # see what's inside
```
The bundle is unsigned. To sign for distribution, run `mcpb sign --self-signed` (smoke-test) or `mcpb sign` with a real cert. See `mcpb sign --help`.
`releases/*.mcpb` is gitignored โ upload the artefact to the GitHub releases page (or wherever your install URL points at) after building.
## Project Structure
```
testingbot-mcp-server/
โโโ src/
โ โโโ index.ts # Entry point
โ โโโ server-factory.ts # MCP server setup
โ โโโ config.ts # Configuration management
โ โโโ lib/
โ โ โโโ types.ts # TypeScript types
โ โ โโโ constants.ts # Constants and enums
โ โ โโโ error.ts # Error classes
โ โ โโโ logger.ts # Logging setup
โ โ โโโ utils.ts # Utility functions
โ โ โโโ get-auth.ts # Authentication helpers
โ โ โโโ app-automate-client.ts # REST client for Maestro/Espresso/XCUITest
โ โโโ tools/
โ โโโ maestro.ts # Maestro run tools (upload, run, poll, diagnose)
โ โโโ app-automate.ts # Espresso & XCUITest run tools
โ โโโ browsers.ts # Browser & device tools
โ โโโ tests.ts # Test management tools
โ โโโ builds.ts # Build management tools
โ โโโ storage.ts # Storage tools
โ โโโ screenshots.ts # Screenshot tools
โ โโโ user.ts # User management tools
โ โโโ team.ts # Team management tools
โ โโโ cdp.ts # Chrome DevTools Protocol tools
โ โโโ tunnels.ts # Tunnel management tools
โ โโโ live.ts # Live testing session tools
โโโ tests/
โ โโโ tools/ # Unit tests
โโโ dist/ # Compiled output
โโโ package.json
```
## Troubleshooting
### Authentication Errors
If you see authentication errors:
1. Run the `tb_login` tool (just ask the agent to "log me in to TestingBot") โ the simplest fix, no key/secret needed
2. Or verify your credentials at https://testingbot.com/members/user/security
3. Ensure environment variables are set correctly (env vars override the `tb_login` credentials file)
4. Check that there are no extra spaces in your credentials
### Connection Issues
If the MCP server won't connect:
1. Restart your MCP client (Claude Desktop, VS Code, etc.)
2. Check the MCP client logs for errors
3. Verify Node.js version is >= 20
### Tool Execution Errors
If tools fail to execute:
1. Check the server logs for detailed error messages
2. Verify you have sufficient permissions/quota in your TestingBot account
3. Ensure all required parameters are provided
### Debug Logging
To enable debug logging, set the `LOG_LEVEL` environment variable:
```bash
export LOG_LEVEL=debug
```
Or in your MCP client configuration:
```json
{
"mcpServers": {
"testingbot": {
"command": "npx",
"args": ["-y", "@testingbot/mcp-server"],
"env": {
"TESTINGBOT_KEY": "your-api-key",
"TESTINGBOT_SECRET": "your-api-secret",
"LOG_LEVEL": "debug"
}
}
}
}
```
## Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
1. Fork the repository
2. Create your feature branch (`git checkout -b feature/amazing-feature`)
3. Commit your changes (`git commit -m 'Add some amazing feature'`)
4. Push to the branch (`git push origin feature/amazing-feature`)
5. Open a Pull Request
## License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
## Support
- ๐ง Email: support@testingbot.com
- ๐ Issues: [GitHub Issues](https://github.com/testingbot/mcp-server/issues)
- ๐ Documentation: [TestingBot MCP Documentation](https://testingbot.com/support/ai/mcp)
## Related
- [TestingBot API Package](https://github.com/testingbot/testingbot-api)
- [Model Context Protocol](https://modelcontextprotocol.io)
- [TestingBot Platform](https://testingbot.com)
TDQS
Scored across 73 tools
Most tools target distinct resources or workflows, and descriptions often clarify family-specific tools (Maestro vs App Automate vs visual vs accessibility). However, there is notable overlap among session-start tools (startLiveSession, startDesktopLiveSession, startMobileLiveSession, tb_openBrowser, createCdpSession), tunnel cleanup tools (stopTunnel vs deleteTunnel), and screenshot-related tools. The large surface makes boundary decisions harder even when each description is informative.
Many tools follow a camelCase verb-noun convention, but the set mixes unprefixed camelCase names, a tb_ snake_case-prefixed browser-control family, and noun-style names like maestroCheatSheet. The tb_ grouping is deliberate and readable, but overall naming is not uniform across the server.
At 73 tools, the server is far beyond the typical 3-15 well-scoped range and is heavy even for a broad testing platform. Several tools are convenience wrappers or narrow variants, which adds surface area without proportional independent capability.
The server covers many TestingBot workflows: sessions, tunnels, storage, tests/builds, Maestro, App Automate, accessibility, visual regression, account, and team. Still, several subdomains lack full lifecycle coverage, such as create/update/delete for accessibility and visual tests, cancelling App Automate runs, and deleting Maestro projects. These are notable gaps but agents can often work around them through the dashboard or existing tools.