MCPControl
An MCP server for Windows: the agent moves the mouse, presses keys, manages windows and takes screenshots
- 332 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent drives the mouse and keyboard of a real desktop and can click or type anything
- Screenshots and the clipboard expose personal data
- A server address over SSE without HTTPS exposes computer control to everyone on the network
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/mcpcontrolDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent drives the mouse and keyboard of a real desktop and can click or type anything
- Screenshots and the clipboard expose personal data
- A server address over SSE without HTTPS exposes computer control to everyone on the network
This is third-party code. Review the repository files before installing.
What it does
MCPControl lets a model drive a Windows desktop. For the mouse there are precise movement, clicks and dragging. Keyboard tools type text, press combinations and hold keys. For windows the server lists them, switches focus and changes size and position. It captures the whole screen or the active window and also works with the clipboard. Under the hood there are three automation providers, switchable with the `AUTOMATION_PROVIDER` variable. The default is keysender, native Windows automation. There is a PowerShell provider for simple operations and an AutoHotkey v2 provider for advanced scripts. Operations can be split across providers, for example the keyboard to AutoHotkey and the clipboard to PowerShell. The author advises running the server in a Windows virtual machine at 1280 by 720, because click accuracy is tuned for that resolution. The client connects to the server over SSE: the server starts with `mcp-control --sse` and the client gets its address.
Who it is for. For people who want Claude to act in Windows apps that have no API: UI testing, data entry, repetitive operations.
Good fit when
- You want a UI test of a desktop app where the agent presses buttons and reports problems
- You need to fill forms and move data between programs that cannot talk to each other
- You want a product demo recorded with the agent performing steps on a live screen
- You want an agent to take over repetitive work in an app with no API while you watch
- You want an experiment where the agent plays a simple game from screenshots
Not a fit when
- You need macOS or Linux: the project supports Windows only
- The task fits an API or a command line: that is faster and more reliable than clicking the screen
- You need accuracy across several monitors: multi-screen functions behave unreliably
Example request
Open Notepad, type a short greeting into it and take a screenshot to show the result.MCP
- Transport
- sse
- Authentication
- not required
Environment variables
AUTOMATION_PROVIDER- Automation provider: keysender, powershell or autohotkey
AUTOHOTKEY_PATH- Path to the AutoHotkey v2 executable when it is not on PATH
Limitations
It works on Windows only. Building the native modules needs Visual Studio Build Tools with the C++ workload, Python and Node.js. Minimizing and restoring windows is unsupported, and multi-screen functions behave unreliably. Some actions need elevated permissions. A click offset and scaling bug remains.
The author calls the project experimental and potentially dangerous and takes no responsibility for data loss or other consequences. Run the server in a controlled environment.
The last commit in the repository dates from December 2025.
How to disable. Stop the mcp-control process and delete the MCPControl entry from the client settings. The package is removed with `npm uninstall -g mcp-control`.
FAQ
How do I connect MCPControl to Claude?
Install the package with `npm install -g mcp-control` and run `mcp-control --sse`. In the Claude settings add an MCPControl server with transport sse and an address like http://192.168.1.100:3232/mcp, using your own IP and port. After the client restarts, the server appears in the MCP menu.
Why does the author recommend a virtual machine?
The server drives the mouse and keyboard of a real desktop. An agent's mistake inside a virtual machine stays there and does not touch your main system.
Which automation providers are supported?
Three: keysender by default, PowerShell for simple operations and AutoHotkey v2 for advanced ones. Set a common provider with `AUTOMATION_PROVIDER`, and keyboard, mouse, screen and clipboard each have separate variables.
Is HTTPS needed for remote access?
Yes, the MCP specification requires HTTPS for network transports in production. Run `mcp-control --sse --https --cert path_to_cert --key path_to_key` and add `--port` if needed.
Why do clicks miss the buttons?
Accuracy is tuned for 1280 by 720 and a single screen. At other scales an offset bug remains that the author has not fixed yet.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for driving Windows: the agent opens apps, clicks and types, reads window and UI state, and works with files and PowerShell
An MCP server for macOS: the agent moves the mouse, types, manages windows and captures the screen when an app has no API
Clawd Cursor: desktop control through MCP
Clawd Cursor
An MCP server that lets an agent drive windows and apps: it reads the screen as structure and acts by element name
MCP server for AppleScript: the agent runs scripts and controls Mac apps, from Notes and Calendar to files
Open Computer Use
open-computer-use
An open MCP server for desktop control: the agent reads app state and clicks, types and scrolls on macOS, Linux and Windows
A Swift MCP server that lets an agent drive any macOS app: click, type and read the interface tree
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MCPControl (this entry) | no | not required | 332 | |
| Windows-MCP | no | not required | 10,740 | also drives Windows, while MCPControl adds a choice of automation provider and an SSE mode for working in a virtual machine |
| Clawd Cursor: desktop control through MCP | no | not required | 406 | an MCP server that lets an agent drive windows and apps |
README in short
The mcp-control package ships through npm, and installing it builds native modules for Windows. Before the first run the author asks you to read the danger warning and the list of known limitations.