mcp-mokowindows

A Model Context Protocol (MCP) server that lets AI assistants like Claude operate a Windows desktop -- audio, display, processes, services, power, network, registry, storage, WSL, winget, plus window capture and full Chrome browser automation. 80+ tools, each backed by Win32 APIs or PowerShell, exposed through the standard MCP tool interface.

Highlights

• Windows system control -- audio devices and per-app volume, displays, processes and services, power, network, firewall, registry, environment, scheduled tasks, storage, and more.
• Occlusion-tolerant window capture (windows_window_capture) -- screenshot a single application or browser window via the Win32 PrintWindow API, even when it is minimized, occluded, on another monitor, or on another virtual desktop.
• Chrome browser automation & testing (windows_chrome_control) -- launch and drive a dedicated, isolated Chrome over the DevTools Protocol: navigate, click, type, run JS, and screenshot -- built for testing dev apps. Includes responsive/mobile viewport emulation (Bootstrap breakpoints + device presets), selectable Chrome profiles, and one-time Google OAuth sign-in.
• Zero heavy dependencies -- only @modelcontextprotocol/sdk and zod; CDP uses Node's built-in fetch and global WebSocket.

Install

npm install -g @mokoconsulting/mcp-mokowindows

Requires Node.js 20+ on Windows (uses Win32 APIs and PowerShell). Google Chrome is needed only for the windows_chrome_control tool.

Register with Claude Code

claude mcp add -s user mokowindows -- mcp-windows

Or add it to your MCP config (~/.claude.json or a project .mcp.json):

{
  "mcpServers": {
    "mokowindows": {
      "type": "stdio",
      "command": "mcp-windows"
    }
  }
}

Restart Claude Code so it loads the server at startup.

Tool categories

| Area | Examples | |---|---| | Core | windows_execute, windows_system_info, windows_terminal_* | | Files | windows_file_read/write/edit, windows_search, windows_drives | | Processes & services | windows_process_list/kill, windows_service_list/control | | Audio & display | windows_audio_*, windows_display_*, windows_theme_* | | Power & network | windows_power_*, windows_network_*, windows_wifi_*, windows_firewall_* | | System admin | windows_registry_*, windows_env_*, windows_task_scheduler_*, windows_startup_* | | Packages & WSL | windows_winget_*, windows_wsl_* | | Capture & browser | windows_window_capture, windows_chrome_control |

See the Wiki for the full tool reference.

Browser automation for testing

windows_chrome_control runs a dedicated Chrome (its own profile, separate from your everyday browser) and is aimed at testing web apps:

action: launch | status | open | navigate | reload | back | forward |
        wait | wait_for_selector | scroll | screenshot | get_text | eval |
        click | type | insert | press | close_tab | quit | google_login

• Responsive screenshots -- screenshot accepts a preset (xs/sm/md/lg/xl/xxl Bootstrap breakpoints, or device presets iphone-se/iphone-14/pixel-7/galaxy-s22/ipad/ipad-pro) or explicit width/height/mobile/device_scale_factor. Captures mobile layouts at true device resolution.
• Reliable when hidden -- captures work even if the window is minimized, occluded, or unfocused.
• Profiles -- profile and user_data_dir params, or set a machine default with windows_mcp_config (chromeProfile, chromeUserDataDir) so every launch uses it.
• Google OAuth -- action:"google_login" signs into Google once in a persistent profile so "Sign in with Google" flows work on later runs (passkeys and Windows Hello supported since it is real headful Chrome).

Configuration

Config lives at ~/.mcp_windows.json and is managed with windows_mcp_config:

| Key | Purpose | |---|---| | chromeProfile | Default Chrome profile (--profile-directory) for windows_chrome_control | | chromeUserDataDir | Default Chrome profile store (--user-data-dir) | | blockedCommands | Commands refused by windows_execute | | allowedDirectories | Filesystem access allow-list | | outputLineLimit, commandTimeout | Output and timeout limits |

Installation

Published to npm as @mokoconsulting/mcp-mokowindows. Requires Node.js 20+ on Windows.

npm install -g @mokoconsulting/mcp-mokowindows

Or run without a global install:

npx @mokoconsulting/mcp-mokowindows

See Install above and the Installation guide for prerequisites (Google Chrome is only needed for the windows_chrome_control tool).

Usage

Register the server with Claude Code and it exposes 80+ Windows system tools to the agent:

claude mcp add -s user mokowindows -- mcp-windows

Or add it to your MCP config (~/.claude.json or a project .mcp.json) as shown in Register with Claude Code. Restart the client so it loads the server at startup. Once connected, the agent can control audio, displays, processes, services, power, network, registry, storage, WSL, and winget, capture individual windows, and drive an isolated Chrome for browser testing. See Tool categories and the Wiki for the full tool reference.

Documentation

Full tool reference and guides are on the Wiki:

• Home -- tool inventory by category
• Installation
• Configuration

Contributing

Contributions are welcome. See CONTRIBUTING.md and follow the MokoStandards.

License

Licensed under the GNU General Public License v3.0 or later -- see LICENSE.

Moko Consulting -- MokoStandards