Zoom Transcript Fetch

A standalone agent skill: fetch Zoom AI Companion transcripts on demand or on a schedule, and turn them into styled meeting notes — automatically, after every call.

No more manual note-taking. No more lost decisions.

Learn more: https://myaifreedomsystems.github.io/zoom-transcript-fetch-site/

What you get

• scripts/zoom_fetch_latest.py — a tested, stdlib-only Python script (no pip installs) that handles Server-to-Server OAuth, meeting-instance resolution, retries while Zoom processes, and dedup so every transcript downloads exactly once.
• styles/ — four pluggable note-style extraction templates (bullet summary, detailed third-person, action items only, coaching tracker). Add your own by dropping in a markdown file with the same shape.
• SKILL.md — the full skill definition in agentskills.io format: the two Zoom transcript systems, the fetch chain, and every pitfall we hit so you don't have to.
• ONBOARDING.md — zero-to-working setup in about 15 minutes: Zoom Marketplace app creation, required account settings, credentials, first test run, and scheduling.

Requirements

• Python 3.9 or later
• A Zoom account with AI Companion enabled (included in paid Zoom plans)
• A Zoom Server-to-Server OAuth app (free to create — see ONBOARDING.md)

Install

1. Clone this repo into your agent's skills directory:

   git clone https://github.com/MyAiFreedomSystems/zoom-transcript-fetch.git

2. Create the Zoom app and enable transcript retention (Steps 1–2 of ONBOARDING.md — about 10 minutes of clicking in the Zoom portal).

3. Configure credentials:

   cd zoom-transcript-fetch
   cp .env.example .env   # fill in the three ZOOM_ values
   chmod 600 .env

4. Test it:

   ZOOM_ENV_FILE=.env python3 scripts/zoom_fetch_latest.py --verbose --once

5. Schedule it with cron, launchd, Task Scheduler, or your agent runtime's scheduler. Examples in ONBOARDING.md.

How it works

After each meeting ends, the script polls Zoom's AI Companion transcript endpoint, downloads new transcripts as .vtt files with speaker labels, and records what it fetched so nothing downloads twice. Your agent then reads each new transcript, applies the note style you chose, and stores formatted notes to Google Docs, Notion, or local markdown.

Full details — including the two separate Zoom transcript systems (and why querying the wrong one silently returns nothing) — are in SKILL.md.

License

MIT