agentsclimarketplace

Capture windows application

Skill NoLegitHere/capture-windows-application-mcp/skills/capture-windows-application

Use the capture-windows MCP to launch Windows executables, discover visible desktop windows, and capture application screenshots. Use when an agent needs to inspect or verify a native Windows app window, capture a running process by PID or title, or take multiple frames from a launched executable during local Windows UI validation.From its SKILL.md

Install
npx -y skills add NoLegitHere/capture-windows-application-mcp --skill capture-windows-application

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.0 KB, 576 tokens by cl100k_base, as published. Nobody here has run it

Capture Windows Application

Use the capture-windows MCP for native Windows desktop screenshots. Prefer it for local .exe windows that browser tools cannot inspect.

Load The MCP

Use the configured MCP server named capture-windows. Harnesses may prefix MCP tool names differently, but this server exposes:

  • list_windows: discover visible desktop windows and their process IDs.
  • capture_window: capture an already-running visible window by process_id or by window_title.
  • run_and_capture: launch an executable, wait for its window, capture one or more screenshots, and optionally close it.

If these tools are unavailable, ask the harness to configure or enable this repo's local MCP server before claiming the window was inspected.

Choose A Tool

  1. Use list_windows first when the target app is already open or the title is uncertain.
  2. Use capture_window with process_id from list_windows whenever possible. Use window_title as a fallback or as a filter inside the selected process.
  3. Use run_and_capture when the task should launch a specific .exe. Pass a full executable_path; add arguments, window_title, and a larger wait_ms when startup or title matching needs it.
  4. Use capture_count and capture_interval_ms for a short visual sequence. The MCP returns the last successful capture inline and reports saved paths for the capture set.
  5. Set close_after only when the launched or reused process should be terminated after capture.

Capture Workflow

For an open app:

  1. List windows.
  2. Select the visible window by process name, title, and PID.
  3. Capture by PID.
  4. Inspect the inline image or the saved PNG path before claiming a visual result.

For an app to launch:

  1. Resolve the executable path from the repo, build output, or known Windows path.
  2. Call run_and_capture.
  3. Increase wait_ms if the process starts before its window becomes visible.
  4. Add window_title when one executable can own multiple visible windows.
  5. Inspect the returned image before reporting visual verification.

Guardrails

  • This MCP captures visible Windows desktop windows. It is not a DOM inspector and cannot validate hidden or headless UI.
  • Prefer a user-provided or workspace-local save_path when the screenshot artifact should stay with the task. Without it, screenshots are saved under the MCP server's .screenshots folder.
  • Treat run_and_capture as potentially reusing an already-running process with the same executable. Review close_after before enabling it.
  • If capture fails because no window appears, use list_windows, inspect the actual title and PID, and retry with a better title filter or longer wait.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,286. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.