Guides and resources
Guides, videos, setup pages and comparisons for giving AI agents real Android phones: to test your app, automate apps without an API, or run inside your product.
56 resources
Start here
- 01Cloud phones for AI agents: what they are and how to choose one.What a cloud phone for an AI agent is, the three kinds on the market, and the checklist that separates agent-ready phones from device farms and multi-account tools.
- 02Test your Android app with an AI agent.Let Claude Code, Codex or Cursor build your APK, install it on a cloud Android phone, walk the screens it changed and report what it saw.
- 03AI QA testing for Android apps: the tools, and how to run it with your own agent.AI QA tools for Android, from managed services to plain-English test runners, compared with the build-it-yourself route: your coding agent, a cloud Android phone, a short task, and a bug report you can check.
What is a cloud phone? How it works, what it costs and who it's for.
A cloud phone is an Android phone that runs in a data center and that you use over the internet, from a browser, an app or an API. The three kinds, what each costs, whether a free one exists, and what a cloud phone can't do.
A phone sandbox for AI agents: let the agent use its own Android phone, not yours.
Giving an AI agent your own phone means your messages, banking apps and accounts are a mistake away. A sandboxed cloud phone holds only what you put on it, takes keys you can limit and revoke, records what the agent did, and wipes back to clean in one call.
AI exploratory testing for Android apps: charters, a worked session, and the report to ask for.
Exploratory testing is where an AI agent beats a script: it decides the next step from what the screen shows. How to write charters an agent can follow, timebox a session on a cloud Android phone, what the agent actually sees and does, and how its findings become regression tests.
How AI agents use mobile apps: reading the screen, choosing an action, and checking what happened.
An AI agent uses a phone app in a loop: read the screen, pick one action toward the goal, act, and read the screen again to check. How agents read Android screens (pixels, the accessibility tree, or both), how they target taps, how they recover from failures, and where the phone can run.
Cloud Android phone APIs in 2026: what each one controls, what it costs, and a whole session in curl.
Nine services let you drive a cloud Android device from code, but they control very different things: some give you ADB, some only run test frameworks, some read the screen for an agent. A comparison of what each API controls, how it bills and what survives between sessions, then the Phonebox API end to end.
Monitor an Android app's screen automatically and send what it shows to your website.
When the number or status you need only exists inside a mobile app, a signed-in cloud phone can check it on a schedule: open the app, read the screen as text, post the value to your webhook, and park. The script, the schedule, what to do when the app signs out, and what each check frequency costs.
How to run Android apps on a Mac in 2026: the emulator, Genymotion, BlueStacks Air, a real phone or a cloud phone.
A Mac can't run Android apps on its own. Five ways to get Android on one, what each needs from your Mac, what it costs, and which fits testing an app you're building, playing a game or letting an AI agent use the phone.
Android emulator for iPhone and iPad: what actually works in 2026.
There's no Android emulator on the App Store, and Apple's rules explain why. What works instead: an Android phone in the cloud that you use from Safari, browser app previews, and consumer cloud-phone apps. What each costs and what each can't do.
How to run an APK online in your browser, and what to do with an .aab.
Run an Android app from your computer's browser without installing an emulator: upload the APK to a cloud Android phone, or install it from the command line. Converting an Android App Bundle first, size limits by service, and why installs fail.
Android emulator in Codex cloud, Claude Code on the web and other cloud agent sandboxes: what works.
The Android emulator needs KVM, and most cloud sandboxes for coding agents don't document it. Which sandboxes can run an emulator, which can't, and how to give a cloud agent a real Android phone over HTTPS instead.
Check your Android app for accessibility issues with an AI agent, and what still needs TalkBack.
What Google's accessibility checks measure and their thresholds, which tools run them, and how an AI agent on a cloud Android phone finds unlabeled controls and duplicate labels on every screen you changed. Plus what only a person with TalkBack can judge.
Appium testing and AI agents in 2026: Appium 3, Appium MCP, and an agent on a real Android build.
Appium is still the standard for scripted Android UI tests across device farms. What changed in Appium 3 and 4, what the official Appium MCP server adds for coding agents, what Appium suites cost to run in the cloud, and where an AI agent on a cloud Android phone fits next to them.
Automate Android UI testing without writing test code: recorders, plain-English tests and AI agents.
Ten ways to automate Android UI tests without writing Espresso or Appium code: Maestro's YAML, recorders, plain-English tools, Google's previews and AI agents. How each is authored, where it runs, what it costs and whether you can export code.
The best Android MCP servers in 2026: twelve compared, local and hosted.
Twelve MCP servers that let an AI agent control Android, compared on what they need, whether a phone comes with them, iOS support and how the agent reads the screen. GitHub stars and activity checked October 16, 2026.
Reproduce Android bug reports with an AI agent, and get a video of the repro.
Paste a bug ticket into Claude Code, Codex or Cursor. The agent installs the build on a cloud Android phone, follows the steps while recording the screen, and hands back the video and the exact steps.
AI test prompts for Android apps: copy-paste tasks for your coding agent.
Five prompts that turn Claude Code, Codex or Cursor into an Android tester: a smoke check, an exploratory session, a bug repro, an accessibility pass and a release check, each on a cloud phone.
Firebase Test Lab is shutting down: where to run your Android tests next.
Google deprecated Firebase Test Lab and will shut it down on September 30, 2027. What moves to Developer Device Platform, what doesn't, and where an AI agent on a cloud phone replaces Robo-style crawling.
What is agentic testing? A practical definition for mobile app teams.
Agentic testing puts an AI agent in the execution loop: it gets a goal, decides each step from what it sees on screen, and reports pass or fail. How it differs from scripted and AI-assisted testing, where it fails, and how to try it on Android.
Mobile app testing tools in 2026: frameworks, device clouds and AI agents.
The Android and iOS testing tools teams use in 2026, by job: frameworks for scripted tests, device clouds to run them, AI QA tools, and AI agents with their own phone. Versions and MCP support included.
BrowserStack alternatives for Android app testing in 2026, with prices and MCP support.
Eight BrowserStack alternatives for Android app testing, compared on entry price, billing unit, MCP access for AI agents and whether a device keeps its state, with the monthly cost of each at three volumes.
Detox testing and AI agents: React Native end-to-end tests, plus an agent on a real Android build.
Detox gives React Native apps fast gray-box end-to-end tests on an emulator. An AI agent on a cloud Android phone checks the release APK the way a user would. What each needs, and how they fit together.
Flutter integration tests, Patrol and AI agents: what to script and what to let an agent check.
Flutter's integration_test and LeanCode's Patrol script end-to-end tests in Dart; Patrol also reaches native UI and ships an MCP server. Where an AI agent on a cloud Android phone fits next to them.
Android emulator online in 2026: the options compared, and a cloud phone that keeps your app.
Ten ways to use Android in a browser, compared on what's free, whether your own APK installs and whether anything is kept between sessions. Then how to test your APK on a Phonebox cloud phone, record it and share a live link.
Appium MCP, mobile-mcp or a hosted Android MCP server: which should your agent use?
Appium MCP, mobile-mcp and Maestro MCP run on your machine and drive a local emulator or USB device; a hosted Android MCP server brings the phone. Setup commands for all four, and which works on a laptop, in CI and in a cloud sandbox.
Maestro testing and AI agents: script the flows you know, let an agent find the rest.
Maestro's YAML flows pin down the journeys you already know. An AI agent on a cloud phone explores the ones you didn't write yet, and can hand each bug back as a new Maestro flow.
AI QA testing for Android apps: the tools, and how to run it with your own agent.
AI QA tools for Android, from managed services to plain-English test runners, compared with the build-it-yourself route: your coding agent, a cloud Android phone, a short task, and a bug report you can check.
What are cloud phones for AI agents?
Cloud phones for AI agents, explained: what they are, why agents need them, how they compare with emulators, device farms and phone numbers, and what to look for.
Give Codex an Android phone (MCP).
Give Codex a real cloud Android phone in 57 seconds: three lines in config.toml, then a real run that reads the screen, taps through an app and parks the phone.
Give Cursor an Android phone (MCP).
Give Cursor's agent a real cloud Android phone in 56 seconds: one entry in mcp.json, then a real run that reads the screen, taps through an app and parks the phone.
Give your AI assistant its own Android phone.
A personal AI assistant can't order groceries or check a delivery in an app that has no API. Give it its own cloud phone, signed in to your apps, and approve payments through a live view link.
Mobile infrastructure for AI agent products: a phone per customer, sign-in handoff and safe failures.
What a product needs when its agents work inside mobile apps: a persistent phone per customer, a screen the model can read, a human handoff for sign-ins, failures the agent can act on, isolation between tenants and costs that stop when nobody is using a phone.
Run Android checks in GitHub Actions without an emulator.
GitHub's hosted runners make the Android emulator slow and flaky. Install each build on a cloud Android phone from the workflow instead, check it, and park the phone.
Test a location-based Android app from any city, with an AI agent.
Set a cloud Android phone's GPS location and the timezone follows. Your agent then checks what the app shows in London, Tokyo or São Paulo, without a VPN or a mock-location app.
Test Android deep links and push notifications with an AI agent.
Let an agent open every deep link in your app, check where each one lands, then open the notification shade and tap your push. It reads the screen as text, so it can tell you exactly what went wrong.
Give Claude Code an Android phone: Android MCP server setup.
Set up Phonebox's hosted Android MCP server in Claude Code, then watch a real run: install an APK, read the screen, tap through an app, hand off to a person, and park.
Give Claude Code an Android phone (MCP).
Watch Claude Code drive a real cloud Android phone in 56 seconds: one command, the screen as text, taps and types, a human handoff, and parking.
How to give Claude Code an Android phone: ADB, an emulator, a local MCP server or a cloud phone.
Five ways to let Claude Code see and control Android: a USB phone with ADB, scrcpy, a local emulator, a local MCP server such as mobile-mcp, or a hosted cloud phone over MCP. Commands for each, where each breaks, and which to pick.
Give Gemini CLI an Android phone: a hosted Android MCP server in one command.
Add Phonebox's hosted Android MCP server with gemini mcp add, and Gemini CLI can read, tap and install apps on a real cloud Android phone.
Give Windsurf an Android phone: an Android MCP server for Cascade.
Add Phonebox's hosted Android MCP server to mcp_config.json and Windsurf's Cascade agent can build, install and test your app on a real cloud Android phone.
Give Cline an Android phone: a hosted Android MCP server for Cline.
Add Phonebox's hosted Android MCP server to cline_mcp_settings.json as a streamableHttp server, and Cline can run and test your Android app on a cloud phone.
Give GitHub Copilot an Android phone: an Android MCP server for VS Code agent mode.
Add Phonebox's hosted Android MCP server to .vscode/mcp.json and Copilot's agent mode can install, read and tap through your Android app on a cloud phone.
Phonebox vs Mobile Next: persistent Android phones for agents or isolated real-device sessions.
Mobile Next gives agents isolated sessions on real iOS and Android devices on minute plans, plus the open-source mobile-mcp. Phonebox gives agents Android phones that keep their apps and sign-ins, billed by the second.
Phonebox vs Appetize: persistent Android phones over MCP or disposable emulator sessions.
Appetize streams iOS and Android emulators that clear when the session ends, on minute plans. Phonebox gives agents Android phones over a hosted MCP server that keep their apps and sign-ins, billed by the second.
Phonebox vs BrowserStack for AI agents: a phone your agent keeps, or a device farm for test suites.
BrowserStack runs QA suites on 35,000+ real devices that are wiped after every session, on annual plans. Phonebox gives AI agents Android phones that keep their apps and sign-ins, billed by the second over MCP.
Give Codex an Android phone: the Android MCP server for OpenAI Codex.
Add Phonebox's hosted Android MCP server to ~/.codex/config.toml and Codex gets a real cloud Android phone: it reads the screen, taps, types and installs your APK.
Give Cursor an Android phone: an Android MCP server for Cursor's agent.
Add Phonebox's hosted Android MCP server to .cursor/mcp.json and Cursor's agent can run, read and tap through your Android app on a real cloud phone.
How an AI agent can use Android apps that have no API.
Five ways an AI agent can get work done in a mobile app, from the app's own API to Android AppFunctions, Gemini, accessibility-based agents and a cloud phone of its own. What each needs, who can use it, and where WhatsApp, food and ride apps draw the line.
Phonebox vs Limrun: persistent Android phones or iOS and Android sandboxes for coding agents.
Both give AI agents cloud Android through MCP. Limrun adds iOS simulators and remote builds on credit plans; Phonebox is Android-only, billed by the second, and keeps phones signed in between sessions.
Cloud phones for AI agents: what they are and how to choose one.
What a cloud phone for an AI agent is, the three kinds on the market, and the checklist that separates agent-ready phones from device farms and multi-account tools.
The best cloud phones for AI agents in 2026, compared.
Phonebox, Mobilerun, Limrun, Mobile Next, Appetize, Genymotion, Multilogin, GeeLark and the big device farms: what each is built for, how agents connect, and what it costs.
A hosted Android MCP server for Claude Code, Cursor and any MCP client.
Connect one URL and your agent gets Android phone tools: create a phone, read the screen, tap and type, install an APK, hand off to a human, park.
Test your Android app with an AI agent.
Let Claude Code, Codex or Cursor build your APK, install it on a cloud Android phone, walk the screens it changed and report what it saw.
Cloud Android emulator or cloud Android phone? Run Android in the cloud by the second.
Skip the local emulator: start an Android phone in the cloud with one API call, install any APK, control it from code or an AI agent, and pay only while it runs.
Phonebox vs Mobilerun: phones billed by the second or a bundled agent platform.
Both give AI agents hosted Android phones through MCP and an API. Phonebox plans are phone time billed by the second; Mobilerun bundles device minutes, its own agent and telephony extras.
A Genymotion Cloud alternative built for AI agents.
Same $0.06 a minute, billed per second instead of per started minute, with persistent phones, a hosted MCP server and a CLI your coding agent can drive.
Give your coding agent an Android device.
Connect Claude Code, Codex or Cursor, run a small first task, and park the phone when you finish.