zoom-video-download

name	zoom-video-download
description	Download the video (MP4) from a Zoom recording share page. "zoom video download", "zoom 録画 ダウンロード", "zoom 動画 ダウンロード" などで起動。
metadata	{"author":"pokutuna","compatibility":"Requires playwright-cli (or playwright MCP), uv, and curl"}
allowed-tools	Bash(uv run *),Bash(playwright-cli *),Bash(curl *),Bash(python3 *),mcp__playwright__browser_navigate,mcp__playwright__browser_snapshot,mcp__playwright__browser_fill_form,mcp__playwright__browser_click,mcp__playwright__browser_evaluate,mcp__playwright__browser_run_code,mcp__playwright__browser_close

Zoom Recording Video Download

Download the MP4 video from a Zoom recording share page.

A Zoom recording is served as one or more single MP4 files (not HLS segments) from ssrweb.zoom.us. The signed URL alone is NOT enough to download — the server also requires the session cookies and a Referer header (any zoom.us value works; the host need not match the recording's region). The flow is therefore:

1. Open the recording in a browser (establishes the authenticated session).
2. Enter the passcode if required.
3. Read the highest-resolution <video> source URL directly from the play page.
4. Export the browser cookies.
5. Download with curl using those cookies + a Referer header (curl streams, resumes, and shows progress for the large file).

Required User Input

Ask the user for:

1. Recording share URL (e.g. https://<region>.zoom.us/rec/share/...)
2. Passcode (if the recording is passcode-protected — some recordings need only the URL)
3. Output file path (where to save the .mp4; default to the recording title)

Primary path: deterministic helper script (playwright-cli)

When playwright-cli is available, run the bundled script — it performs the entire flow (navigate → passcode → URL extraction → cookie export → curl download) deterministically in one command:

uv run --script ${CLAUDE_PLUGIN_ROOT}/skills/zoom-video-download/scripts/download_zoom_video.py \
  --url '<recording_share_url>' \
  --output '<output_file_path>' \
  --passcode '<passcode>'   # omit if the recording has no passcode

Notes:

• Pass --passcode as a single argument; the script JSON-encodes it internally, so shell-special characters in the passcode are handled safely. Single-quote it so the shell does not interpret #, !, $, etc. (e.g. --passcode '#jZ4R==v').
• Run it in the foreground and wait for it to finish. The download is several hundred MB and can take 1–2 minutes; use a generous command timeout. Do not background it and report before it exits — wait for the exit code. The script prints curl progress to stderr and the final saved path to stdout. It supports resume (curl -C -); re-running continues an interrupted download, and re-running on an already-complete file is a no-op success (exit 0). Transient curl errors are retried automatically (3 attempts).
• --keep-open leaves the browser open (useful for debugging); by default the browser is closed as soon as the video URL and cookies are extracted — including on failure (wrong passcode etc.), so no stray browser is left behind.
• The script verifies the output is a non-empty MP4 before exiting. Exit codes: 0 success, 2 passcode required but not given, 3 no <video> source found, 4 download failed (curl error / HTTP non-2xx / empty or non-MP4 output), 5 passcode incorrect.

Verify afterwards that the output is a valid MP4 (e.g. ffprobe).

If playwright-cli is not installed: npm install -g playwright-cli (and uv for the script).

Alternative path: playwright MCP (manual steps)

If only the playwright MCP server is available (no playwright-cli), perform the same flow with MCP tools. The helper script's logic, step by step:

1. Navigate

Call mcp__playwright__browser_navigate with the recording share URL. If the page title is "パスコードが必要" / "Passcode required", do step 2; otherwise skip to step 3.

2. Enter passcode

1. mcp__playwright__browser_snapshot — find the passcode input and submit button refs
2. mcp__playwright__browser_fill_form — fill the passcode
3. mcp__playwright__browser_click — click "レコーディングを視聴" / "Watch recording"

The page then navigates to .../rec/play/....

3. Read the highest-resolution video URL

Call mcp__playwright__browser_evaluate with this function (no need to press Play). A recording usually has two <video> elements — ..._as_<W>x<H>.mp4 (high-res shared screen) and ..._avo_<W>x<H>.mp4 (low-res speaker overlay); pick the largest by width:

() => {
  const vids = Array.from(document.querySelectorAll('video'));
  const scored = vids.map(v => {
    const u = v.currentSrc || v.src;
    const m = u && u.match(/_(\d+)x(\d+)\.mp4/);
    return { url: u, w: m ? parseInt(m[1], 10) : 0 };
  }).filter(x => x.url);
  scored.sort((a, b) => b.w - a.w);
  return scored[0] ? scored[0].url : '';
}

If empty, the player may still be loading (retry), or the recording is audio-only / download-disabled.

4. Export cookies

Call mcp__playwright__browser_run_code with this function and save the returned text to a file (e.g. /tmp/zoom_cookies.txt):

async (page) => {
  const cookies = await page.context().cookies();
  const lines = ['# Netscape HTTP Cookie File'];
  for (const c of cookies) {
    const includeSub = c.domain.startsWith('.') ? 'TRUE' : 'FALSE';
    const secure = c.secure ? 'TRUE' : 'FALSE';
    const expires = c.expires && c.expires > 0 ? Math.floor(c.expires) : 0;
    lines.push([c.domain, includeSub, c.path || '/', secure, expires, c.name, c.value].join('\t'));
  }
  return lines.join('\n');
}

5. Download with curl

UA='Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36'
curl --fail -C - -o '<output_file_path>' \
  -b /tmp/zoom_cookies.txt \
  -H "Referer: https://zoom.us/" \
  -H "User-Agent: $UA" \
  -w 'DONE http=%{http_code} size=%{size_download} time=%{time_total}s\n' \
  '<video_url>'

6. Clean up

Call mcp__playwright__browser_close; remove the temp cookie file.

Troubleshooting

• curl returns 403: cookies or the Referer header are missing/expired. Re-export cookies (step 4) while the browser session is still open, and confirm Referer is set. The signed URL is NOT self-authenticating — both cookies and a Referer are required.
• No <video> element / empty URL: the player may still be loading (retry), or the recording is download-disabled or audio-only.
• Want the speaker overlay too: each <video> is a separate file (_as_ = shared screen, _avo_ = small speaker overlay). Download both URLs if needed; mux with ffmpeg only if you specifically need them combined.
• Signed URL expired mid-download: the URL has a time limit (DateLessThan in its policy). If resume fails, re-run to fetch a fresh URL, then resume.
• Video looks small / very low bitrate: this is usually normal. The _as_ track is a screen/slide share; for mostly-static content (slides, code) even a 1080p recording can be a few hundred kbps. That is the source bitrate, not a wrong-track selection — there is no higher-quality source to fetch (Zoom serves only the _as_ and _avo_ files, no adaptive/HLS variants).