Your agent sees and uses an Android device.
Some problems only show up on a real phone: a page that breaks at phone width, an app that crashes on start, a sound that never plays. MCP Scrcpy lets your agent use an Android phone, tablet or emulator the way you would. It looks at the screen, taps, swipes, scrolls and types, records the screen with its sound, installs and opens apps, and reads what the device logged.
Nothing needs to be installed on the device first.
What you gain over scripting by hand.
A script can send taps to a phone, but it cannot see the screen and cannot tell whether anything happened. An agent that can look, act and look again does the work a person does at the device.
Nothing to install on the device
No app appears on the device
- Typical alternative
- A remote-control app on the phone, with its own permissions and consent dialog
- Here
- A tiny helper file is copied to the device's temporary folder and run from there
The agent can see the screen
It taps where the button really is
- Typical alternative
- Blind taps at guessed positions
- Here
- The agent gets a picture of the screen, acts on what it sees and checks the result with a second picture
One agent at a time
A second agent cannot tap in the middle of the first one's test
- Typical alternative
- Two scripts fighting over the same screen
- Here
- One conversation takes exclusive control of a device and lets go when it ends
Evidence you can replay
A repro you can play back
- Typical alternative
- A screen recorder started by hand
- Here
- The screen is recorded to a video file on your machine, with the phone's sound when it can be captured, or as a WebM that skips repeated frames
Apps and logs in one place
Install, run, look, read the log and clean up without leaving the conversation
- Typical alternative
- Separate commands for installing, logs and removing
- Here
- Installing, opening, reading logs and removing apps happen in the same conversation as the taps
Phones, tablets and emulators
The same way of working for a real phone and a test emulator
- Typical alternative
- A tool for one device type
- Here
- Anything your computer can reach, over USB, Wi-Fi or as an emulator
Ready in the image, ADB is the one prerequisite.
The service comes with the MAIT image and updates with it (mait-image). It needs Android's standard connection tools on the same machine and a device that has approved that machine. It does not start an emulator, pair a Wi-Fi device or accept the approval prompt for you.
Looking at the screen and WebM recordings need a video decoder from Cisco, which is downloaded the first time unless you supply one or forbid the download. Decide that before first use. Plain recording does not need it.
You can also place it behind MCP Registry, so one entry in your agent's app reaches it.
Where it sits in the MAIT toolkit.
It is the screen and touch for your agent on Android, where the other tools cover desktops.
Where to read on.
- Get started, Use cases and Tools