Teams Transcript to Markdown

About

Capture the transcript from a Microsoft Teams meeting recording and download it as clean, LLM-friendly Markdown — roughly 75% fewer tokens than the raw WebVTT file (measured with tiktoken's cl100k_base and o200k_base tokenizers on real Teams recordings). Or grab the .vtt itself, your call.

────────────────────────────────────────
WHAT IT DOES
────────────────────────────────────────

Click the toolbar icon on a Teams recording page, click Capture, done. The popup downloads a Markdown file with the transcript and shows a result panel with Open and Show-in-folder buttons so you can jump straight into the file. No menus, no settings, no setup. Your last-used preferences (format, timestamps, speaker merge) persist between uses.

────────────────────────────────────────
TWO CAPTURE PATHS — PICKED AUTOMATICALLY
────────────────────────────────────────

• SharePoint fast path (~1–3 seconds)
  On *.sharepoint.com/.../stream.aspx pages, the extension reads the SharePoint drive and item IDs from the page's bootstrap data, calls the REST endpoint for the underlying transcript, and downloads the actual .vtt file Microsoft already has on the server. Two HTTP calls. Done.

• DOM-scrape fallback (~20–40 seconds)
  On other recording surfaces (legacy Stream, *.cloud.microsoft, etc.), the extension scrolls the rendered transcript pane and harvests entries from the DOM. Slower, but works without a JSON API. Same Markdown output.

────────────────────────────────────────
LLM-FRIENDLY OUTPUT
────────────────────────────────────────

The Markdown is token-optimized:

• Header block with title, source URL, duration, and ordered speakers
• One paragraph per turn: [MM:SS] Speaker: text
• Consecutive turns from the same speaker get merged into a single paragraph
• Trailing/leading word-overlap between adjacent cues gets trimmed
• Cue IDs, millisecond timestamps, and <v Speaker> voice tags are stripped

Measured on two real 2-hour Teams meeting recordings: ~61% smaller in bytes, ~78–79% fewer tokens with both the cl100k_base (GPT-3.5/4/4-turbo) and o200k_base (GPT-4o/4.1/o1) tokenizers. The token savings come from the fact that VTT files are full of tokenize-hostile artifacts — UUIDs, fractional-second timestamps, voice tags — that BPE tokenizers fragment into many small tokens.

────────────────────────────────────────
RAW WEBVTT OPTION
────────────────────────────────────────

If you'd rather post-process the original file yourself, switch the popup's "Save as" radio to "Raw WebVTT" and you'll get the millisecond-precise .vtt as Microsoft has it on the server. Fast-path only.

────────────────────────────────────────
PRIVACY
────────────────────────────────────────

• No telemetry, no analytics, no servers owned by the extension author
• No host permissions
• No background service worker
• The extension activates only when you click it
• It talks only to the SharePoint origin you are already viewing, using your existing browser session
• Nothing leaves your machine that wasn't already going to SharePoint

────────────────────────────────────────
OPEN SOURCE
────────────────────────────────────────

MIT-licensed. Not affiliated with Microsoft.

Repository:    https://github.com/bkrabach/teams-transcript-md
Project site:  https://bkrabach.github.io/teams-transcript-md/
Privacy:       https://bkrabach.github.io/teams-transcript-md/privacy/