ControlMCP
Specifically supports IntelliJ IDEA with workflows for starting, running configurations, and observing logs.
Integrates with JetBrains IDEs for run configuration selection, run-panel switching, log monitoring, and shortcut automation.
Specifically supports PyCharm with workflows for running configurations and monitoring logs.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ControlMCPPlan a desktop workflow: open Chrome, search for MCP servers"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ControlMCP
😆You’re already a mature LLM, so you should learn to operate the computer by yourself. 🛠️MCP server for LLM-controlled computer operations — screen capture, window management, mouse & keyboard automation.
Overview
ControlMCP is a Model Context Protocol (MCP) server that gives LLMs the ability to see and control a computer — take screenshots, manage windows, move/click the mouse, type on the keyboard, and chain all of these into complex automation workflows.
The repository also ships with a reusable agent skill at skills/computer-control/. It packages desktop-operation SOPs, shortcut guidance, JetBrains IDE workflows, and screenshot-to-click coordinate rules for agents that support skills.
Related MCP server: helix-pilot
Quick Start
Installation
install from source:
git clone https://github.com/nix18/ControlMCP.git
cd ControlMCP
pip install -e .Launch
control-mcpThe server communicates over stdio (standard MCP transport). Configure your MCP client to connect to the control-mcp command.
MCP Client Configuration
Add to your MCP client config (e.g. Claude Desktop, Cursor, etc.):
{
"mcpServers": {
"control-mcp": {
"command": "control-mcp",
"args": []
}
}
}Tools (34 total)
Control Plane
Tool | Description |
| Convert a vague desktop instruction into a structured plan |
| Run a structured plan through the guarded executor |
| Query the current status of a high-level execution run |
| Explicitly approve or reject a sensitive action |
| Rebuild context after shortcut misuse or UI drift |
| Persist reusable workflow experience |
Screen Capture
Tool | Description |
| Full screen or monitor screenshot |
| Region screenshot (x, y, width, height) |
| Stitch a long screenshot while scrolling inside a fixed region |
| List all monitors with resolution |
| Read a screenshot file as Base64 text |
| Convert a grid cell + anchor into precise screen coordinates |
| Resolve screenshot grid metadata and click directly |
Window Management
Tool | Description |
| List all visible windows |
| Find windows by title substring |
| Bring a window to the foreground |
| Focus + screenshot a specific window |
Mouse Control
Tool | Description |
| Click at coordinates (single/double/multi/hold) |
| Drag from point A to point B |
| Move cursor without clicking |
| Get current cursor position |
| Scroll wheel up/down |
Keyboard Control
Tool | Description |
| Press keys or hotkey combinations |
| Hold keys for a duration |
| Type text character by character |
| Execute a timed sequence of key actions |
Combined Operations
Tool | Description |
| Execute a mixed sequence of mouse + keyboard + wait + screenshot actions |
Additional Actions
Tool | Description |
| Get clipboard text |
| Set clipboard text |
| Launch an application |
| Open a URL in the browser |
| Pause for N seconds |
| Get RGB color at screen coordinates |
| Press a keyboard shortcut |
Examples
See docs/TUTORIAL.md for comprehensive usage examples.
// Plan a vague desktop task first
{"tool": "plan_desktop_task", "args": {"instruction": "Switch to PyCharm and run the current config"}}
// Execute a generated plan
{"tool": "execute_desktop_plan", "args": {"plan_id": "plan_abc123"}}
// Take a screenshot
{"tool": "capture_screen", "args": {}}
// Take a sharper screenshot when text clarity matters
{"tool": "capture_window", "args": {"title": "PyCharm", "quality": 75, "sharpen": true}}
// Read that screenshot as Base64 text for non-multimodal models
{"tool": "read_screenshot_base64", "args": {"file_path": "/tmp/screen.jpg"}}
// Click at (500, 300)
{"tool": "mouse_click", "args": {"x": 500, "y": 300}}
// Combined: click → select all → type
{"tool": "mouse_and_keyboard", "args": {"actions": [
{"action": "click", "x": 500, "y": 300},
{"action": "key_press", "keys": ["ctrl", "a"]},
{"action": "key_type", "text": "New text"}
]}}Rebuilt Workflow
ControlMCP now supports a control-plane-first workflow for higher precision desktop automation:
Normalize the user instruction with
plan_desktop_taskReview or directly execute the structured plan
Let the guarded executor choose a faster observation strategy (
capture_window/capture_region/capture_scroll_region)Verify each critical step and recover when context is lost
Require explicit confirmation for payment/password/asset-related actions
Save successful workflow experience for future runs
For small or visually ambiguous targets, you can also ask capture_screen, capture_region,
or capture_window to generate a second grid_file_path overlay image with grid_rows and
grid_cols, then convert a chosen cell + anchor through resolve_grid_target before clicking.
Documentation
Document | Description |
This file | |
Chinese version of this file | |
Requirements analysis | |
Architecture design | |
Module design | |
Functional design | |
Tutorial & examples | |
Agent Skill: computer operation SOPs | |
Skill-specific install and usage guide | |
Window rescue and window shortcut reference | |
JetBrains IDE run/log observation workflow |
Agent Skill
The skills/computer-control/ folder contains a ready-to-use Agent Skill that teaches LLMs how to operate computers proficiently.
What is included
SKILL.md: the main skill instructions, SOPs, shortcut tables, and common failure patternsdocs/coordinate-system.md: coordinate conversion reference for screenshot-to-click workflowsdocs/window-management.md: window maximize/restore/snap shortcuts and window recovery workflowdocs/idea-run-workflow.md: JetBrains IDE startup, run-panel switching, and log stabilization workflowREADME.md: skill-local installation and usage notes
What the skill covers
Keyboard-first automation: prefer shortcuts over UI clicking whenever possible
Plan-before-act control plane: normalize ambiguous instructions before touching the desktop
Window recovery: fix minimized, half-screen, or partially restored windows before further actions
Coordinate-safe clicking: convert screenshot-local coordinates into screen coordinates explicitly
IDE workflows: IntelliJ IDEA / PyCharm run-configuration selection, run-panel switching, and log monitoring
Sensitive-action gating: require confirmation before payment/password/asset-related steps
Operational fallback: when JetBrains shortcuts do not behave as expected, check the local
ReferenceCard.pdfor JetBrains official documentation
Install the skill into your agent
You can either copy skills/computer-control/ into your agent's skill directory, or add it via a symbolic link.
Option 1: copy the directory
# Codex CLI
cp -r skills/computer-control ~/.codex/skills/
# Claude Code
cp -r skills/computer-control ~/.claude/skills/
# OpenCode
cp -r skills/computer-control ~/.config/opencode/skills/Option 2: create a symbolic link
On macOS / Linux:
# Codex CLI
ln -s "$(pwd)/skills/computer-control" ~/.codex/skills/computer-control
# Claude Code
ln -s "$(pwd)/skills/computer-control" ~/.claude/skills/computer-control
# OpenCode
ln -s "$(pwd)/skills/computer-control" ~/.config/opencode/skills/computer-controlOn Windows (Command Prompt as Administrator when required):
mklink /D "%USERPROFILE%\.codex\skills\computer-control" "%CD%\skills\computer-control"
mklink /D "%USERPROFILE%\.claude\skills\computer-control" "%CD%\skills\computer-control"
mklink /D "%USERPROFILE%\.config\opencode\skills\computer-control" "%CD%\skills\computer-control"Using a symbolic link is convenient while iterating on the skill, because changes in this repository are reflected immediately in the agent's skills directory.
If your agent supports custom skill paths, you can also reference this folder directly.
Use the skill
After installation, invoke it naturally in prompts such as:
Use $computer-control to restart the IDEA app and wait until logs stop updatingUse $computer-control to maximize the target window and capture itUse $computer-control to operate PyCharm with keyboard shortcuts first
For skill-specific details, see skills/computer-control/README.md.
Project Structure
ControlMCP/
├── README.md # This file
├── README.zh-CN.md # Chinese README
├── LICENSE # GNU GPLv3 license
├── pyproject.toml # Package config
├── src/
│ └── control_mcp/
│ ├── __init__.py
│ ├── server.py # MCP server + tool registration
│ ├── schemas/
│ │ ├── __init__.py
│ │ └── responses.py # Structured response types
│ ├── tools/
│ │ ├── __init__.py
│ │ ├── screen.py # Screen capture tools
│ │ ├── window.py # Window management tools
│ │ ├── mouse.py # Mouse control tools
│ │ ├── keyboard.py # Keyboard control tools
│ │ ├── combined.py # Combined operations
│ │ └── actions.py # Additional actions
│ └── utils/
│ ├── __init__.py
│ ├── capture.py # Capture utilities (JPEG, resize)
│ ├── _win_window.py # Windows backend
│ ├── _mac_window.py # macOS backend
│ └── _linux_window.py # Linux backend
├── skills/
│ └── computer-control/ # Agent Skill: computer operation SOPs
│ ├── SKILL.md # Main skill instructions
│ ├── docs/
│ │ ├── coordinate-system.md # Coordinate system reference
│ │ ├── window-management.md # Window management reference
│ │ └── idea-run-workflow.md # JetBrains IDE run/log workflow
│ └── README.md # Skill install & usage guide
├── docs/
│ ├── REQUIREMENTS.md
│ ├── ARCHITECTURE.md
│ ├── MODULE_DESIGN.md
│ ├── FUNCTIONAL_DESIGN.md
│ ├── TUTORIAL.md
│ └── zh-CN/ # Chinese documentation
│ ├── REQUIREMENTS.md
│ ├── ARCHITECTURE.md
│ ├── MODULE_DESIGN.md
│ ├── FUNCTIONAL_DESIGN.md
│ └── TUTORIAL.md
└── tests/
├── __init__.py
├── test_schemas.py # 22 tests
├── test_screen.py # 6 tests
├── test_window.py # 11 tests
├── test_mouse.py # 13 tests
├── test_keyboard.py # 16 tests
├── test_combined.py # 12 tests
└── test_actions.py # 13 testsPlatform Support
Platform | Screen Capture | Window Management | Mouse/Keyboard |
Windows | ✅ mss | ✅ pygetwindow | ✅ pyautogui |
macOS | ✅ mss | ✅ Quartz | ✅ pyautogui |
Linux | ✅ mss | ✅ xlib | ✅ pyautogui |
License
GNU General Public License v3.0 (GPLv3)
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/nix18/ControlMCP'
If you have feedback or need assistance with the MCP directory API, please join our Discord server