MAS Agent
MAS Agent: the Android AI agent that turns one sentence into a macro you run forever
Type what you want done in plain English. MAS Agent, the AI agent for Android automation, explores the live app on an emulator like BlueStacks, LDPlayer, MuMu or MEmu, on an always-on cloud device, or on your own Android phone plugged in, captures every button and counter it needs from the real screen, and compiles a deterministic Python macro validated by replay.
Then the model leaves the loop: every run costs zero AI credits, on as many devices as you like. No root, no ADB scripting, no APK modification.
- 0 credits per run
- 3-day free trial
- Emulators, Mac, cloud devices
- MCP for Claude Code, Cursor, Codex and more
Try a prompt
Four things people ask MAS Agent to do.
youClaim the daily reward, then close any popups that show up.
MAS AgentOn it. Exploring the app now, macro coming up.
youCheck the delivery app every 10 minutes and accept any new order under 5 km.
MAS AgentReading the distance with OCR before each accept. Runs every 10 minutes on the cloud device.
youOpen the seller app each morning, read the sales from yesterday for each of our 8 stores, and send them to my webhook.
MAS AgentOne exploration on the first store, then the same macro across all 8 accounts. Totals posted to your webhook.
youTest the login flow and send me the results.
MAS AgentDriving the flow with your test account. Results posted when it passes.
Watch Macro Automation Studio in action: the studio, the agent, and a macro running on a device.
Cost per run
Zero AI cost per run: the model leaves the loop
Most agents keep a model in the loop on every run. You pay per minute or per token forever, each action waits on a model call, and the same task can go differently tomorrow. MAS spends credits once, while the agent explores and compiles. The macro it hands you replays deterministically, at script speed, with no model in the loop. Your hundredth run is as cheap as it is predictable.
Explore once, run forever
How the Android AI agent works: describe, explore, harvest, compile, validate, run forever
Six steps, one exploration. A mobile AI agent does the first five; a plain macro does the sixth, for as long as you like.
-
1. Describe
One sentence in plain English. The agent turns it into a plan with a goal, an end condition and a budget you confirm before anything is spent.
-
2. Explore
It drives the live app the way a person would: screens, buttons, popups, counters. When it needs a decision it stops and asks instead of guessing.
-
3. Harvest
Every template image and OCR region it will use is captured from the real screen and checked on the device before it goes into the code.
-
4. Compile
The exploration becomes a standard MAS Python project in house style, an AI macro maker for games and apps: bounded loops, recovery preambles, decisive checks.
-
5. Validate
The macro is replayed through the unmodified SDK until it passes 3 clean runs in a row. Only then does it graduate into Studio.
-
6. Run forever
Schedule it, hand it an argument form, run it across a device group or on a cloud device with your PC off. No model in the loop, so every run is free and behaves the same way.
The agent also keeps what it learned. App maps (screens, proven routes, keyframes) are stored as memory, so the next task in the same app starts from what the last one found.
Built for
Game farmers, multi-account operators, developers and QA
-
Game farmers
Daily rewards, farming loops, event check-ins, rally and hunt routines: built once, then run every day on an emulator or on a cloud device with your PC off.
-
Multi-account operators and teams
Explore on one device, run on a device group. On the Business plan an organization shares macros, devices and cloud devices across seats, with owner, admin and member roles.
-
Developers and QA
An editable Python project, the full SDK, and an MCP server for Claude Code, Cursor, Codex, Hermes or any MCP client. Drive a login or checkout flow with a test account and get a pass or fail line back, with no test framework to maintain.
Games and apps
An AI that plays mobile games for you
The agent builds the routine and the macro runs it. Games are the most common ask, and anything else on the screen works the same way.
Point MAS Agent at the daily grind and let it auto play the boring part: rewards, chests, farming loops, event check-ins, the checklist you repeat on every account. It learns the flow on your emulator or on a cloud device, popups included, because it met those popups on the live screen. The routine is then yours to run overnight, and on a cloud device it keeps going with your PC off. If you have been looking for an AI to farm mobile games while you sleep, this is the shape of it, minus the babysitting. The same goes for any app on the device: a daily checklist in a productivity app, a form to fill, a login flow to test. If it is on the screen, the agent can learn it.
-
Daily rewards and chests
Log in, claim, close whatever pops up, log out. The agent met the popups on the live screen, so the macro knows the way past them.
auto play -
Resource farming loops
Farm on a loop with a stop condition read off the screen: energy at zero, storage full, a timer not yet ready.
auto farm -
Event and timer check-ins
Collect event points every few hours while you sleep. On a cloud device it keeps going with your PC off.
AFK -
Rally, hunt and scavenge routines
Read counters and timers with OCR, act when they are ready, skip when they are not. No coordinates to break.
OCR -
Multi-account checklists
Explore once on one device, then schedule the macro across a device group so every account gets the same routine.
device groups -
App test flows
Drive a login or checkout flow with a test account, assert each screen, and post a pass or fail line you can read from anywhere.
QA
No automation tool is 100% risk-free, so automate responsibly and at your own discretion.
Where it runs
No root, no ADB scripting: emulators on Windows or Mac, cloud devices, or your own phone
MAS Agent is an AI agent for Android phones and for the emulators that stand in for them. It works from the screen with image and text recognition and humanized input, on any device MAS can see and drive: BlueStacks, LDPlayer, MuMu Player and MEmu on Windows or macOS, MAS cloud devices that run server-side, or your own Android phone plugged into the studio. No APK is modified, no root is needed, and there is no ADB scripting for you to maintain. Want phone automation AI that controls an Android emulator? That is the default setup.
- BlueStacks
- LDPlayer
- MuMu
- MEmu
- Windows
- Apple Silicon Mac
- Cloud devices
- Your own phone
Phones gained built-in agents this year: supervised, limited to a few app categories, and bound to the one phone in your hand. None of them run on an emulator, a group of cloud devices or the phone on your desk, and none were built for games. MAS Agent was.
You stay in charge
Steer it live. Set the budget before it starts.
-
It asks instead of guessing
When two readings of the screen are possible, or a step would spend or destroy something, the agent stops and puts the question to you with the choices spelled out.
-
Redirect it mid-run
The composer never locks. Type a new instruction while the agent explores and it lands in the very next decision, no restart needed.
-
A budget, set first
Before a single credit is spent, the plan states its budget and you confirm it. Usage is visible while it works and you can stop at any moment.
For developers
A Python project you can edit, and an MCP tool that authors macros
What the agent hands you is a normal MAS project: app.py, an argument form in script_args.py, and a task file with the schedule and retry settings. Open it in the IDE, change anything with the full Python SDK, and it is still the same macro the scheduler and device groups run.
If you already live in Claude Code, Cursor, Codex, Hermes or any other MCP client, the MAS MCP server exposes the same thing as a tool: author a macro from a prompt, poll the exploration, answer its question, stop it. The same server gives your client the device itself: lease it, see its screen, read text with OCR, find an image, tap, swipe, type, run a macro, fetch the app map. Call it a phone use agent, a mobile use agent, or computer use for Android; the output is the same editable project.
Agent tools on the MCP server
On 2026-08-20 a one-sentence prompt became a graduated macro. Six of six validation runs came back clean, about 15 seconds each, on an unmodified SDK.
A first live test, not a benchmark.
AI credits
Spent while it builds, never while it runs
- 1,000 credits = $1. Credits are spent only when MAS Agent or other AI features run. A finished macro runs at zero credits.
- A budget is set before the agent starts, and it asks rather than guesses when a step would cost more.
- A cloud device includes 20,000 AI credits every month, and runs the finished macro with your PC off.
- Plans start at $5/mo with a 3-day free trial.
credits per run, after the agent has authored the macro. Validated by 3 clean replays first.
Android AI agent FAQ
What is an Android AI agent?
An Android AI agent, also called a mobile AI agent, is an AI that operates Android apps through the screen: it reads the UI, decides what to tap, and acts. MAS Agent goes one step further. Instead of staying in the loop forever, it explores once and compiles what it learned into a deterministic macro you rerun at zero AI cost.
Can AI control my Android phone from my PC?
MAS controls Android apps on an emulator (BlueStacks, LDPlayer, MuMu, MEmu, on Windows or Apple Silicon Mac), on an always-on cloud device, or on your own Android phone plugged into the studio, all driven from your PC or Mac. See cloud devices.
Can an AI agent play mobile games for me?
Yes, with a division of labor. MAS Agent explores the game and builds the macro; the macro then plays on your emulator or a cloud device without the agent in the loop, overnight if you like. No automation tool is 100% risk-free, so automate responsibly and at your own discretion.
What does "explore once, run forever" mean?
Credits are spent only while the agent explores and builds. The output is a standard MAS Python macro whose template images and OCR regions were captured from the live screen. Every later run is deterministic replay: zero credits, no model in the loop, the same behavior every time.
Does MAS Agent need root or ADB?
No root and no APK modification. MAS drives the app through the screen with image recognition, OCR and humanized input, so there is no ADB scripting for you to write. Cloud devices are hosted by MAS and need nothing installed at all.
Which apps and games does the agent work with?
Anything you can run on a supported device: apps and games on BlueStacks, LDPlayer, MuMu Player or MEmu on Windows or Mac, whatever is installed on a MAS cloud device, and the apps on your own phone when you plug it in. The agent works from the screen, so it never depends on an app exposing an API.
How many AI credits does one exploration use?
It depends on how long the agent works: a short daily routine costs less than a long multi-screen flow. Before the agent spends anything you set a budget, and when it is unsure it asks rather than guesses. You can watch usage live and stop at any time. A validated macro then runs at zero credits forever. Rates are on the pricing page.
Does the agent run in the cloud or on my PC?
Both. Point it at an emulator or a plugged-in phone and it explores your local device. Point it at a cloud device and everything runs server-side, so you can close the laptop and check back later. See cloud devices.
Is there a free trial?
Yes. Download Macro Automation Studio and start a 3-day free trial. After the trial, the Individual plan starts at $5/mo. Full details are on the pricing page.
Is there an AI that can play games for you?
Yes. MAS Agent is an AI that plays mobile games for you by learning the routine on an emulator or a cloud device and turning it into a macro. The macro then handles the daily grind: rewards, farming loops, event check-ins, multi-account checklists. If you would rather start from a finished bot, the make a bot for any game page and the marketplace are the shortcut.
Can Claude control my Android phone or emulator?
Through the MAS MCP server, yes. Any MCP client, including Claude Code, Claude Desktop, Cursor, Codex and Hermes, gets 25 tools for Android emulators and cloud devices, including one that starts a MAS Agent exploration and returns the finished macro. See the Android MCP server.
What happens when the game or app updates its screens?
The macro works from template images and text regions captured from the live screen, so a small visual change usually still matches. When a screen changes enough to break a step, you run a new exploration instead of hand-patching coordinates, and the result is validated by replay again before it graduates.
Can I steer the agent while it runs, and edit the macro afterwards?
Yes to both. While the agent explores you can type a new instruction at any moment and it lands in the agent's next decision. When it needs a choice it stops and asks. The finished macro is a normal MAS Python project, open in the IDE with the full SDK, so you can change anything the agent wrote.
Is MAS Agent a phone use agent or a macro recorder?
Both halves, in order. During exploration it is a phone use agent: it sees the screen and acts on it. What it hands you is a macro, but unlike a macro recorder it does not replay coordinates. It finds elements by image and text recognition, recovers from popups it met during exploration, and is validated by 3 clean replays before you get it.