Android-MCP by CursorTouch: Control an Android Phone from Claude, Cursor and Other Apps

• Reads the phone screen: what is on it and where to tap

  Returns the interactive elements on the current screen with their labels and tap coordinates. With the vision option it adds a screenshot, with boxes drawn on it unless you turn that off.

  Try: “What is on my phone screen right now?”

  Read · Snapshot

• Lists the phones and emulators connected over ADB

  Shows each device ADB can see and its state.

  Try: “Which Android devices are connected?”

  Read · ListDevices

• Connects to one device by its serial number

  Opens a connection to the ADB device with the serial you give.

  Try: “Connect to emulator-5554.”

  Write · ConnectDevice

• Lists, connects or disconnects a device

  One tool with three actions: list the devices, connect to one, or disconnect the current one. Wi-Fi targets get port 5555 added when it is missing.

  Try: “Disconnect from the phone.”

  Write · Device

• Taps the screen at given coordinates

  Sends a tap at the x and y position you give.

  Try: “Tap the middle of the screen.”

  Write · Click

• Taps a button or field found by its text or id

  Finds an element by text, resource id, class or description and taps it. It is more reliable than coordinates when the layout shifts, and an index picks the nth match.

  Try: “Tap the button labelled Sign in.”

  Write · ClickBySelector

• Presses and holds at given coordinates

  Sends a long press at the x and y position you give.

  Try: “Long-press the app icon at the top left.”

  Write · LongClick

• Swipes from one point to another

  Drags a finger from the first position to the second, for scrolling or paging.

  Try: “Scroll down the list.”

  Write · Swipe

• Drags something from one spot and drops it on another

  Picks up at the first position and drops at the second.

  Try: “Drag the slider all the way to the right.”

  Write · Drag

• Types text into a field

  Types the text at the position you give, and can clear the field first.

  Try: “Type hello into the search box.”

  Write · Type

• Presses a device button such as Back or Home

  Presses a named key on the device, for example back or volume up.

  Try: “Go back to the home screen.”

  Write · Press

• Pulls down the notification shade

  Opens the notification bar so the assistant can read what is shown there with the next snapshot.

  Try: “Show me my notifications.”

  Write · Notification

• Pauses for a number of seconds

  Sleeps for the number of seconds you give, then returns. It changes nothing on the phone, but the source marks it destructive like the touch tools, so it is not counted as read-only.

  Try: “Wait three seconds, then look again.”

  Write · Wait

• Waits until a button or text appears

  Waits for an element to show up on screen and returns it, or an error on timeout. Better than a fixed wait when content loads slowly.

  Try: “Wait until the Continue button appears.”

  Read · WaitForElement

Watch out for

• It can tap and type in any app on the device

  Click, Swipe, Drag, Type and Press act on whatever is on screen, such as messages, banking apps or settings. Use an emulator or a test device without real accounts. github.com

• Screen contents go to your AI assistant

  Snapshot returns on-screen text and, with the vision option, a screenshot. Anything visible on the phone, including private messages, reaches the model. github.com

• The README warns about untrusted prompts

  It says to use controlled environments such as emulators or test devices when running untrusted prompts or agents. github.com

Security advisories