MCP Desktop Automation
Automate desktop actions like mouse control, keyboard input, and taking screenshots.
MCP Desktop Automation
A Model Context Protocol server that provides desktop automation capabilities using RobotJS and screenshot capabilities. This server enables LLMs to control mouse movements, keyboard inputs, and capture screenshots of the desktop environment.
Configuration to use Desktop Automation Server
Here's how to configure Claude Desktop to use the MCP Desktop Automation server:
NPX
{
"mcpServers": {
"desktop-automation": {
"command": "npx",
"args": ["-y", "mcp-desktop-automation"]
}
}
}
Permissions
This server requires system-level permissions to:
- Capture screenshots of your screen
- Control mouse movement and clicks
- Simulate keyboard input
When first running Claude Desktop with this server, you may need to grant these permissions in your operating system's security settings.
Limitations
While this server works with various MCP clients, it has been primarily tested with Claude Desktop.
Important: The current implementation has a 1MB response size limit. For screen captures, this means:
- High-resolution screenshots may exceed this limit and fail
- Testing has shown 800x600 resolution works reliably
- Consider reducing screen resolution or capturing specific screen areas if you encounter issues
Requirements
- Node.js (>=14.x)
Components
Tools
-
get_screen_size
- Gets the screen dimensions
- No input parameters required
-
screen_capture
- Captures the current screen content
- No input parameters required
-
keyboard_press
- Presses a keyboard key or key combination
- Inputs:
key(string, required): Key to press (e.g., 'enter', 'a', 'control')modifiers(array of strings, optional): Modifier keys to hold while pressing the key. Possible values: "control", "shift", "alt", "command"
-
keyboard_type
- Types text at the current cursor position
- Input:
text(string, required): Text to type
-
mouse_click
- Performs a mouse click
- Inputs:
button(string, optional, default: "left"): Mouse button to click. Possible values: "left", "right", "middle"double(boolean, optional, default: false): Whether to perform a double click
-
mouse_move
- Moves the mouse to specified coordinates
- Inputs:
x(number, required): X coordinatey(number, required): Y coordinate
Resources
The server provides access to screenshots:
-
Screenshot List (
screenshot://list)- Lists all available screenshots by name
-
Screenshot Content (
screenshot://{id})- PNG images of captured screenshots
- Accessible via the screenshot ID (timestamp-based naming)
Key Features
- Desktop mouse control
- Keyboard input simulation
- Screen size detection
- Screenshot capabilities
- Simple JSON response format
License
This MCP server is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License. For more details, please see the LICENSE file in the project repository.
เซิร์ฟเวอร์ที่เกี่ยวข้อง
Kone.vc
ผู้สนับสนุนMonetize your AI agent with contextual product recommendations
MIST
An AI assistant server for managing notes, Gmail, Calendar, Tasks, and Git.
n8n MCP Server
Automate workflows with n8n, a powerful workflow automation tool.
Project Handoffs
Manages AI session handoffs and tracks next steps for projects.
esa.io
An MCP server for the esa.io API, allowing you to search, read, create, update, and delete posts.
PowerPoint Translator
Translate PowerPoint files using AWS Bedrock. Requires AWS credentials to be configured.
C++ Excel Automation
A C++ based MCP server for intelligent Excel automation using the OpenXLSX library.
Foreman MCP Server
MCP server for Foreman host management and infrastructure automation.
Notion
Manage and interact with your entire Notion workspace.
congressgov-mcp-server
U.S. congressional data
WxO Agent MCP
Simple MCP (Model Context Protocol) server that invokes a single Watson Orchestrate agent remotely. The agent is defined once via environment variables or MCP config. Use this when you want a lightweight MCP that only chats with one agent—no tool management, no agent listing, no flows. Just invoke_agent(message) and get_agent().