Zoom Transcript Fetch · Meeting Notes, Written While You Work

After every Zoom call, clean meeting notes arrive. Nobody wrote them.

A free, open skill that teaches your AI agent to fetch the transcript after each meeting, apply the note style you chose, and file the notes where your team keeps things. You stay on the call. The agent does the paperwork.

Get the Skill, Free on GitHub

See Example Notes

Works with any agent that can run a Python script and follow a SKILL.md. No subscription. No new app to maintain.

Decisions get made on calls. Then they disappear.

You finish a client call and the real work starts: writing up what was said, who agreed to what, and what happens next. Skip it and the decisions evaporate. Do it and you lose half an hour you already spent once. Either way it costs money, time, and energy, and it repeats after every single meeting.

STEP 01

The call ends. The fetch begins.

On a schedule you set, the skill checks your Zoom account for new AI Companion transcripts. It handles the OAuth, the retries while Zoom processes, and the bookkeeping so every transcript downloads exactly once. Missed runs catch themselves.

STEP 02

Your style gets applied

The agent reads the raw transcript and rewrites it in the note style you picked: bullet summary, detailed third-person, action items only, or a coaching tracker that carries threads between sessions. Quotes stay verbatim. Deadlines are never invented.

STEP 03

Notes land where you keep things

Google Docs, Notion, or plain markdown files next to the transcripts. Every note is named with date, topic, and style so it sorts itself, and links back to the raw transcript so every line is auditable.

One repo. Everything tested on real calls.

The skill is a SKILL.md your agent reads, a stdlib-only Python script with zero pip installs, four note-style templates, and an onboarding guide that takes about fifteen minutes, most of it clicking through the Zoom portal.

The fetch script

Downloads the transcripts

Stdlib-only Python. Nothing to install, nothing to break
Server-to-Server OAuth with token handling done for you
Retries while Zoom processes, up to 40 minutes
Dedup state file: every transcript downloads exactly once
Silent when there is nothing new, loud only on real errors

The note styles

Turns words into notes

Bullet Summary for fast internal reads
Detailed Third-Person for clients who missed the call
Action Items Only for standups and weekly reviews
Coaching Tracker for recurring advisory calls
Add your own style by dropping in one markdown file

github.com/MyAiFreedomSystems/zoom-transcript-fetch →

What setup actually costs

No invented before-and-after stats here. These are the mechanics of the skill itself, and you can check every one of them in the repo.

15 min

from clone to first fetched transcript

pip installs, the script is stdlib-only

note styles included, yours added in one file

documented pitfalls the script already handles

The half hour after every call, returned.

Count the recurring calls on your calendar. Each one ends with the same tax: someone writes the notes, or nobody does and the decisions live in memory until they don't. The skill removes the tax. Money you stopped spending on write-ups. Time back after every meeting. Energy for the work the meeting was actually about.

The notes are also better than the ones written from memory. They come from the full transcript, with verbatim quotes for every commitment and a source link for every line. When a client says "that's not what we agreed," you have the sentence where the agreement was made.

Free, open, and ready for your agent.

Clone the repo into your agent's skills folder, spend fifteen minutes in the Zoom portal, and run one command. The onboarding guide walks every step.

1. Grab the skill git clone https://github.com/MyAiFreedomSystems/zoom-transcript-fetch.git 2. Follow ONBOARDING.md Create a free Zoom Server-to-Server OAuth app, enable transcript retention, paste three values into a .env file. 3. Test it, then schedule it ZOOM_ENV_FILE=.env python3 scripts/zoom_fetch_latest.py --verbose --once

Then tell your agent which note style to use and where to file the notes. Cron, launchd, Task Scheduler, or your agent runtime's own scheduler all work. Read the README for the full setup.