What is RobotActions Device Farm?

RobotActions Device Farm is a cloud testing platform that gives you real Android and iOS devices, Apple TV and Google TV boxes, and real browsers, in our cloud. Reach them from your browser, drive them with Selenium, Appium, Playwright or WebdriverIO, or let an AI agent drive them over MCP.

How many devices can I test on simultaneously?

Your concurrency is simply the number of devices on your plan: buy 11 devices and you can run 11 sessions at once. Sessions and test minutes on those devices are unlimited. A device is yours alone for the session and goes back to the pool when you release it; every device is restored to a clean state overnight. Enterprise plans get a dedicated host and devices that keep their state.

Do you provide real devices or emulators?

Real physical devices, always — no emulators and no simulators. They behave differently from the hardware your users actually hold, which is the whole reason to use a device farm.

What integrations do you support?

RobotActions works with Jenkins, GitHub Actions, GitLab CI/CD and other CI providers, and with Selenium, Appium, Playwright and WebdriverIO. Over MCP, an AI agent can also reach Jira, TestRail and Azure DevOps — reading stories and acceptance criteria, updating test cases and filing defects. Our API allows custom integrations with your existing workflow.

Can I record my test sessions?

Yes, all test sessions can be recorded with our video playback feature. This allows you to review test executions, analyze failures, and share recordings with your team for collaborative debugging.

Can I test geo-restricted applications?

Yes for location itself — you can emulate the device GPS position and mock API responses, so location-based features in your own app are fully testable on a real device. One limit worth stating plainly: we do not control the IP address your traffic leaves from, so a restriction a third party enforces by source IP is out of scope today.

Yes. Apple TV (tvOS) and Google TV / Android TV boxes are in the device list alongside phones, at the same per-device price. Live control gives you a full remote in the browser — d-pad, OK, back, home, play/pause, volume, power, and typing straight into the TV — next to a live stream with audio as well as video, so you can check audio and playback, not just the screen. AI agents drive them over MCP, and Appium suites run on them the same way they run on a phone.

Do live sessions include audio?

Yes. Live sessions stream audio as well as video on every device type — Android phones and tablets, iPhone and iPad, Apple TV, Google TV / Android TV, and desktop browsers. Test video and music playback, notification sounds, in-app audio cues and media controls with the device's real audio, not just the screen.

Can I test with VoiceOver or TalkBack on a real device?

Yes, on both platforms: VoiceOver on iPhone and iPad, TalkBack on Android. Turn the screen reader on from the panel in a live session and drive it with Previous, Next, Activate, Back, Home and Read All, or with the arrow keys and Enter (plus the rotor on iOS). No gestures to memorise. The panel shows the exact text the screen reader speaks for each element as a running transcript, and the spoken output streams to your computer with the rest of the device audio. Missing labels, a wrong reading order and unhelpful announcements show up in writing, not only in the audio. Switch the device language, or only the app's, to check it in every language your app ships in.

Can I run accessibility audits on real devices?

Yes. Run an accessibility scan on a real iPhone, iPad or Android device and get an audit report of what it finds — from the console, or let your AI agent run the same scan over MCP and file what it finds in Jira, Azure DevOps or TestRail. Pair it with the screen-reader panel (VoiceOver or TalkBack) to check how the screen actually reads, and with device or app language switching to repeat the check in every language you ship.

MCP AI Agent Integration

Learn how to connect AI agents like Claude and Cursor to real Android and iOS devices using Model Context Protocol (MCP). Complete setup guide with code examples.

The server gives your agent tools. The skills give it judgement.

Once connected, your agent has hundreds of tools and no guidance on sequencing them. Our free, open-source Agent Skills cover which tool, in what order, and the mistakes that quietly turn a passing test into one that tested nothing.

Browse Agent Skills

Add RobotActions as a connector (OAuth)

Add RobotActions to your AI assistant as a remote connector and sign in with your RobotActions account. No token to copy and nothing to install. Use this server URL everywhere:

Server URL

https://mcp.robotactions.com/mcp

Claude (claude.ai and Claude Desktop)

In Claude's connector settings, add a custom connector with the URL above, then sign in with your RobotActions account when prompted.

ChatGPT

With developer mode or custom connectors enabled on your plan, create a connector with the same URL and sign in with your RobotActions account.

Grok

Add it as a custom connector at grok.com/connectors with the same URL. If a header is requested instead of sign-in, use an API token from Settings → API Tokens as Authorization: Bearer <token>. Grok setup is per user: each person adds the connector to their own account.

Prompts to try once connected

“Open https://example.com in Safari on an iPhone and tell me which network requests fail.”
“Set the device location to Paris, open Maps and take a screenshot.”
“Turn on VoiceOver on an iPad and read me the home screen.”
“Install my build from Firebase App Distribution and check the login screen for missing accessibility labels.”
“File a Jira bug for what you found, with steps and the screenshot.”

To disconnect, remove the connector in your assistant. API tokens can be revoked in Settings → API Tokens. Config-file clients and CI keep working with API tokens, as described in the tabs below.

What is MCP?

Model Context Protocol (MCP) is an open protocol that enables AI applications to securely connect to external data sources and tools. RobotActions supports MCP integration, allowing AI agents like Claude to directly control real mobile devices.

Why use MCP with RobotActions?

AI-Powered Testing: Let AI agents write, execute, and debug tests autonomously

Natural Language Control: Describe what you want to test in plain English

Intelligent Debugging: AI analyzes failures and suggests fixes

Automated Exploration: AI discovers edge cases you might miss

Prerequisites

A RobotActions account with an API key
Claude Desktop, Cursor IDE, or Claude Code CLI installed
Node.js 18+ for running the MCP server