---
tags:
  - zoom-ai-transcript-export
  - personal-os
  - automation
  - zoom
  - pgvector
  - webhook
---
# Automating Zoom AI Meeting Transcript/Summary Export

Связано: [[executor-security-analysis]]

Yes — multiple paths, listed fastest to most flexible.

## Options

### 1. Zoom MCP connector for Claude (no code)
Zoom ships an official MCP connector available in Claude's connector directory. Retrieves summaries, transcripts, recordings, and related meeting content. Use on-demand or via a Cowork workflow.

### 2. Zoom REST API + Server-to-Server OAuth (real automation)
Relevant endpoints:
- `GET /v2/meetings/meeting_summaries` — list all AI Companion summaries
- `GET /v2/meetings/{meetingUUID}/meeting_summary` — fetch a specific summary body
- `GET /v2/meetings/{meetingId}/transcript` — full transcript

Required scopes:
- `meeting:read:list_summaries:admin`
- `meeting:read:summary:admin`
- `meeting:read:summary:master` (sometimes needed for cross-account — currently buggy per active dev forum thread)

Run as a cron/launchd job, poll for new summaries, write to disk / Obsidian / pgvector store.

### 3. Webhook-driven (push)
Zoom emits `meeting.summary_completed` events. Point webhook at your VPS, receive event, call API to pull body. Cleaner than polling.

### 4. Manual (not really automation)
In Zoom Docs: open summary → ⋯ menu → Export to Word/PDF. Per-meeting only.

## Caveats

- Transcripts have limited retention in Zoom — pull regularly, not on-demand months later.
- Summary endpoint OAuth scopes are currently fragile; expect debugging.
- Account admin must enable AI Companion API access; host must have transcript retention on.
- Avoid lethal-trifecta: if pulling via MCP connector in a Claude session, don't combine with web fetch in the same conversation.

## Recommended path

Given existing infra (PostgreSQL + pgvector knowledge base, VPS with reverse tunnel, orchestrator agents):

**Option 3 (webhook) → pgvector** is highest-leverage. Webhook handler on the VPS receives `meeting.summary_completed`, calls the summary + transcript endpoints, writes to pgvector for retrieval by your knowledge agent.

Fallback: Option 2 (polling) if webhook setup is blocked by network/firewall.