SCREENCAST-MCP / CI

Record test runs in CI

Browser tests get a video of every failure for free. Desktop apps on Windows (Electron, WPF, WinUI, WinForms, games) don't. screencast-mcp record runs your test command while it records the runner's desktop, then turns a failure into a GIF and a contact sheet you can open from the run.

00:00THE COMMAND LINE

The same tools the MCP server exposes, without an MCP client. The command's exit code passes straight through.

npx -y @tmhs/screencast-mcp record --target full -- npm run test:e2e

--target: full (default), monitor:0, window:exe=MyApp.exe, or region:x,y,w,h.
--artifacts failure|always|never: when to make evidence.gif and contact-sheet.png (default failure).
--last 30: the GIF covers the last 30 s, where the failure usually is; 0 keeps it all.
--out, --fps, --no-gif, --no-contact-sheet, --discard-on-success, --max-duration.

Everything lands in screencast-evidence/ with an evidence.json summary. Any single tool works the same way: npx -y @tmhs/screencast-mcp call doctor, or call pipeline @recipe.json.

00:10THE ACTION

A composite action wraps it for windows-latest: it checks Node, installs ffmpeg if it is missing, records the command, uploads the evidence folder as an artifact, writes a job summary, and then fails the step with the command's exit code.

jobs:
  ui-tests:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
      - uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6.5.0
        with:
          node-version: 22
      - run: npm ci
      - uses: TMHSDigital/screencast-mcp@main # pin a release tag or SHA
        with:
          command: npm run test:e2e
          target: full          # or window:exe=MyApp.exe
          artifacts: failure    # GIF and contact sheet only when it fails

Outputs: exit-code, gif, contact-sheet, recording, and evidence-dir, for a later step that comments on the PR. All inputs are in action.yml.

00:20WHAT TO KNOW

Hosted Windows runners have an interactive desktop, so gdigrab records what your app draws. This repo records its own capture smoke test with the action on every change to it.
A recording holds whatever is on screen. Keep secrets off the runner's desktop, and remember artifacts are visible to anyone who can read the run.
For a longer suite, keep the default discard-on-success so green runs upload only evidence.json.