Your agent sees a desktop and operates it like a person.
Some things can only be done by looking at a screen and clicking: an installer with no other way in, an application with no automation, a virtual machine that has booted. MCP VNC lets your agent do that. It looks at the screen, moves and clicks the mouse, types text and key combinations, and records a video of what happened.
It works with anything that shows a screen and offers VNC screen sharing: a virtual machine console, a Linux desktop, a Mac. Nothing has to be installed on that desktop, and one agent can work with several desktops at once.
What you gain over scripts and manual remote control.
An application with no automation, a machine with no helper installed and an installer screen all have one thing in common: a screen. Driving the screen is the one way in that always exists, and a recording of it is evidence a person can review.
Any desktop that offers screen sharing
A new machine, a virtual machine console or a Mac is reachable straight away
- Typical alternative
- Install a remote-control helper on every target
- Here
- The agent connects with the address and password you already have, and nothing is installed on the target
It works on what is on screen
The agent can reach an application that has no other way in
- Typical alternative
- A script that needs an interface or an automation hook
- Here
- Looking, clicking and typing work on any application, including ones that expose nothing else
Tuned to the target
Fewer lost keystrokes, and a clear message before typing starts when a text is too long
- Typical alternative
- One typing speed for everything, so keys get dropped
- Here
- Ready-made settings for ordinary desktops, Macs and virtual machine consoles set the typing pace and limits
A recording as evidence
You review what the agent did instead of trusting its summary
- Typical alternative
- Screenshots taken by hand
- Here
- The whole interaction can be recorded as a video, or as a short version with only what changed
Waiting without checking again and again
It stops spending a look on every check
- Typical alternative
- A new look at the screen every few seconds
- Here
- The agent can be told when the screen has stopped changing
Several desktops at once
The agent works across machines without mixing them up
- Typical alternative
- One session per target
- Here
- Every desktop has its own name, up to 16 at a time, and every request says which one it means
Ready in the image, one command to start.
The service comes with the MAIT image and updates with it (mait-image). You give the agent a target: its address and a password, or an account name for a Mac's Screen Sharing. It needs only a network route to that desktop and no internet access. On macOS and Windows, MCP VNC runs in the managed Linux environment that AIDC Launch provides.
You can also place it behind MCP Registry, so one entry in your agent's app reaches it.
Screen sharing carries no sound, so recordings are video only. When the sound belongs in the evidence, use MCP RDP with a target that supports it.
Where it sits in the MAIT toolkit.
It is the screen, mouse and keyboard for your agent on desktops that offer screen sharing.
Where to read on.
- Get started, Use cases and Tools