MCP Application & IDE Screenshot Server

A Windows-focused Model Context Protocol (MCP) server for capturing application windows, IDEs, monitors, and screen regions for visual debugging. It uses PrintWindow for background capture and MSS for foreground screen capture. Capture is fast—typically tens of milliseconds—but depends on the selected mode, window type, and image format.

Install and run

The project targets Python 3.10+ and can be run directly with uvx:

uvx --from . mcp-app-screenshot

For local development:

uv sync
uv run mcp dev server.py

Related MCP server: Screenshot MCP Server

MCP tools

• list_open_windows() lists permitted, visible application windows.

• capture_application_window(target="", mode="background", max_width=1568, max_height=1568, format="png", quality=80, save_path=None, allow_foreground_fallback=False, hwnd=None, pid=None, client_area_only=False, delay_ms=0) captures a chosen window. hwnd, then pid, take priority over target.

• capture_ide_and_app(ide_target="Code.exe", app_target="...", max_width=1568, max_height=1568, format="png", quality=80, allow_foreground_fallback=False) captures both independently, retaining a successful capture when the other fails.

• capture_screen(monitor=0, max_width=1568, max_height=1568, format="png", quality=80, delay_ms=0) captures all monitors (0) or one monitor (1..N).

• capture_region(left, top, width, height, max_width=1568, max_height=1568, format="png", quality=80, delay_ms=0) captures an arbitrary desktop rectangle.

• wait_for_window(target, timeout_s=10) waits for an allowed visible window and returns its description.

• trigger_sharex_workflow(workflow_name="ActiveWindow") triggers one approved ShareX workflow.

Capture tools return a JSON metadata string followed by an MCP image. Metadata includes the window title, process, PID, HWND, original and final dimensions, and the mode used. Supported response formats are png, jpeg, and webp; quality applies to JPEG and WebP.

Security and configuration

The server does not expose windows belonging to password managers by default. These environment variables configure its boundaries:

For example, to expose only VS Code and a development application:

$env:ALLOWED_PROCESSES = "Code.exe,MyApp.exe"
$env:SCREENSHOT_DIR = "D:\mcp-screenshots"

Client configuration

Use uvx from an MCP client configuration:

{
  "mcpServers": {
    "app-screenshot": {
      "command": "uvx",
      "args": [
        "--from",
        "C:\\path\\to\\mcp-app-screenshot",
        "mcp-app-screenshot"
      ],
      "env": {
        "SCREENSHOT_DIR": "C:\\path\\to\\screenshots",
        "ALLOWED_PROCESSES": "Code.exe,MyApp.exe"
      }
    }
  }
}

Limitations

PrintWindow cannot reliably capture popup menus and some GPU-rendered windows. Use mode="foreground" for those applications, or set allow_foreground_fallback=True to opt into bringing the target window to the front. Foreground capture can include anything visually covering the window at capture time.

Demo

Screenshot/GIF demo placeholder.

Development

uv run --with pytest --with ruff pytest
uv run --with ruff ruff check .

License

MIT