Windows Driver Input MCP (Standalone)

Standalone MCP server exposing driver-level keyboard/mouse tools via IbInputSimulator.

No UIA tree, no vision, no shell — just input
Backends: DLL first (ibsim-dll, no AHK required); optional AHK (ibsim-ahk, requires AutoHotkey v2 64-bit)
Supports Unicode text, WASD, combos, holds, and scroll

Repository: https://github.com/win10ogod/windows-driver-input-mcpLicense: MIT

Run

stdio: uv run main.py --transport stdio
SSE: uv run main.py --transport sse --host localhost --port 8001

Configure via env (defaults shown):

WINDOWS_MCP_INPUT_BACKEND=ibsim-dll (or ibsim-ahk)
WINDOWS_MCP_INPUT_DRIVER=AnyDriver
WINDOWS_MCP_RATE_MOVE_HZ=120, WINDOWS_MCP_RATE_MAX_DELTA=60, WINDOWS_MCP_RATE_SMOOTH=0.0
WINDOWS_MCP_RATE_CPS=8.0, WINDOWS_MCP_RATE_KPS=12.0
WINDOWS_INPUT_LOG_LEVEL=INFO
IBSIM_DIR optionally to point to IbInputSimulator directory if not colocated

Tools

Input-Info
Move-Tool
Click-Tool
Drag-Tool
Scroll-Tool
Type-Tool(text, method=unicode|clipboard|vk, press_enter=false)
Shortcut-Tool
Key-Tool(mode=tap|down|up|hold, key, times, interval_ms, hold_ms)
Combo-Tool(keys=[...], hold_ms=...)
Desktop-Info (virtual screen origin/size, monitor count)
Window-Info (active window title/class and rect)
Windows-List (enumerate windows; top-level or children via parent_hwnd; filters: query, only_visible, include_minimized, include_cloaked, limit)
Windows-Select (select by index or first match by query; supports parent_hwnd and visibility flags; returns hwnd and details)
Windows-Activate (bring to front; optional show=restore|minimize|maximize|show, topmost=true|false)
Windows-SetPos (move/resize; loc=[x,y], size=[w,h], z=topmost|notopmost|top|bottom)
Windows-Close (send WM_CLOSE)
Input-RateLimiter-Config

IbInputSimulator

Place IbInputSimulator/Binding.AHK2/IbInputSimulator.dll and IbInputSimulator.ahk either:

Inside this folder (input_driver_server/IbInputSimulator/Binding.AHK2/...), or
Set IBSIM_DIR to the directory holding Binding.AHK2 (or its parent).

The server auto-discovers common locations and also checks the parent repo layout, but for a truly standalone package, vendoring IbInputSimulator in this folder is recommended.

Install

Requirements
- Windows 7–11
- Python 3.13+ (python --version)
- UV package manager: pip install uv (or see https://github.com/astral-sh/uv)
- IbInputSimulator is already vendored in input_driver_server/IbInputSimulator.
- Optional (only when using ibsim-ahk): AutoHotkey v2 64‑bit
Setup (clone once)
- git clone https://github.com/win10ogod/windows-driver-input-mcp.git
- cd windows-driver-input-mcp
- (Optional) uv sync if you prefer a local venv with pinned deps

Client Config Snippets (JSON)

Use these JSON blocks in clients that accept Model Context Protocol server configs.

Claude Desktop (Advanced > Add Local MCP)

Paste this JSON into the “Install from JSON” dialog:

{
  "command": "uv",
  "args": [
    "--directory",
    "<ABSOLUTE PATH TO>/windows-driver-input-mcp",
    "run",
    "main.py"
  ],
  "env": {
    "WINDOWS_MCP_INPUT_BACKEND": "ibsim-dll",
    "WINDOWS_MCP_INPUT_DRIVER": "AnyDriver",
    "WINDOWS_MCP_RATE_MOVE_HZ": "120",
    "WINDOWS_MCP_RATE_MAX_DELTA": "60",
    "WINDOWS_MCP_RATE_SMOOTH": "0.0",
    "WINDOWS_MCP_RATE_CPS": "8.0",
    "WINDOWS_MCP_RATE_KPS": "12.0",
    "WINDOWS_INPUT_LOG_LEVEL": "INFO"
  }
}

Or install as an Extension package:

cd windows-driver-input-mcp
npx @anthropic-ai/mcpb pack
In Claude Desktop: Settings → Extensions → Advanced Settings → Install Extension → pick the generated .mcpb.

Codex CLI (TOML)

Edit %USERPROFILE%/.codex/config.toml and add:

[mcp_servers.windows-driver-input]
command = "uv"
args = [
  "--directory",
  "<ABSOLUTE PATH TO>/windows-driver-input-mcp",
  "run",
  "main.py"
]
env = {
  WINDOWS_MCP_INPUT_BACKEND = "ibsim-dll",
  WINDOWS_MCP_INPUT_DRIVER = "AnyDriver",
  WINDOWS_MCP_RATE_MOVE_HZ = "120",
  WINDOWS_MCP_RATE_MAX_DELTA = "60",
  WINDOWS_MCP_RATE_SMOOTH = "0.0",
  WINDOWS_MCP_RATE_CPS = "8.0",
  WINDOWS_MCP_RATE_KPS = "12.0",
  WINDOWS_INPUT_LOG_LEVEL = "INFO"
}

Save and restart Codex CLI to load the new server.

Gemini CLI (JSON)

Edit %USERPROFILE%/.gemini/settings.json and merge the following under mcpServers:

{
  "mcpServers": {
    "windows-driver-input": {
      "command": "uv",
      "args": [
        "--directory",
        "<ABSOLUTE PATH TO>/windows-driver-input-mcp",
        "run",
        "main.py"
      ],
      "env": {
        "WINDOWS_MCP_INPUT_BACKEND": "ibsim-dll",
        "WINDOWS_MCP_INPUT_DRIVER": "AnyDriver",
        "WINDOWS_MCP_RATE_MOVE_HZ": "120",
        "WINDOWS_MCP_RATE_MAX_DELTA": "60",
        "WINDOWS_MCP_RATE_SMOOTH": "0.0",
        "WINDOWS_MCP_RATE_CPS": "8.0",
        "WINDOWS_MCP_RATE_KPS": "12.0",
        "WINDOWS_INPUT_LOG_LEVEL": "INFO"
      }
    }
  }
}

Restart Gemini CLI to apply the configuration.

Acknowledgements

Portions of functionality are adapted from Windows-MCP:https://github.com/CursorTouch/Windows-MCP
Special thanks to IbInputSimulator for driver-level input:https://github.com/Chaoses-Ib/IbInputSimulator

Windows Driver Input MCP (Standalone)