teams-transcript-to-markdown

Capture a Microsoft Teams meeting transcript from its Stream recording page, or from the meeting's recap in Teams, and download it as clean Markdown, with consecutive speaker turns collapsed and a completeness check that flags missing entries and suspicious time gaps.

Teams renders the transcript as a virtualized list: only the rows near the viewport exist in the DOM at any moment, and off-screen rows are destroyed as you scroll. You can't just select-all and copy. The userscript scrolls the whole list itself, keeps every row it has seen (keyed by data-list-index, so re-rendered rows are de-duped), and converts the result to Markdown in the browser.

Install

1. Install Violentmonkey or Tampermonkey.
2. Open the raw file on GitHub and let the manager offer to install it.

The script's @updateURL points at that same raw file, so the manager picks up new versions on its own (or on demand via its "check for updates"). It only installs one whose @version is higher than the installed one — bump @version in every change you push, or nobody gets it. GitHub caches raw files for about five minutes, so a check straight after a push can still see the old version.

It runs on https://*.sharepoint.com/*stream.aspx* — the page a Teams recording opens in, e.g. https://liveuclac-my.sharepoint.com/personal/…/_layouts/15/stream.aspx?id=…Meeting%20Recording.mp4.

It also runs on https://*.sharepoint.com/*xplatplugins.aspx*, the SharePoint iframe that Teams' Meeting Recaps view embeds. That is the only place a meeting that was transcribed but not recorded still shows its transcript, and the iframe can't be opened in a tab of its own. Userscript managers match each frame against its own URL, so it doesn't matter whether the Teams page around it is teams.microsoft.com or teams.cloud.microsoft. The iframe passes only the file's GUID (?uniqueId=…), so the title, start time and author are looked up through SharePoint's REST API.

Use

1. Open the recording and open its Transcript panel, or open the meeting under Meeting Recaps in Teams and pick its Transcript tab.
2. A Transcript → Markdown button appears bottom-right once transcript rows are on the page.
3. Click it. It scrolls the transcript top to bottom (the button shows captured/total), then downloads <recording name>.md.

The button reports ⚠ N rows missing if the list never rendered some rows; the Markdown is saved anyway and its gap check says where. The userscript manager's menu has the same action, plus Download captured transcript HTML for the raw capture.

Without a userscript manager, or on a page the @match doesn't cover, paste the whole file into the DevTools console instead: it adds the same button and exposes TG.run(), TG.status(), TG.downloadMarkdown() and TG.downloadHTML(). If the transcript panel sits inside an iframe, pick that frame in the console's context dropdown (top-left) first.

What the Markdown contains

• a header with the meeting's metadata: title and recording start (both parsed from the recording's file name, which Teams writes as <title>-YYYYMMDD_HHMMSS-Meeting Recording.mp4), duration (from the video player), who recorded it (from SharePoint's REST API, if you may read the file's details), a link back to the recording, and every speaker in the transcript;
• speaker names and the timestamp each turn started at;
• consecutive segments from the same speaker collapsed into one block, carrying the name forward across continuation rows that have no header;
• system events (joins, screen-sharing, …) as blockquotes;
• a gap / completeness check at the end that reports entries missing from the DOM (the definitive check for a virtualized list) and timestamp jumps over one minute, noting whether a jump spans missing rows or is just one long turn.

Converting a saved HTML capture

The converter needs no browser, so HTML saved with Download captured transcript HTML can be turned into Markdown with Node, since the userscript exports its converter when required:

node -e 'process.stdout.write(require("./teams-transcript-to-markdown.user.js").toMarkdown(require("fs").readFileSync("transcript.html", "utf8")).markdown)' > transcript.md

The header then has only the speakers, since the rest of the metadata comes from the live page.