Files
obsidian-vault/personal/projects/personal-os/zoom-ai-transcript-export.md
T
2026-05-30 07:03:10 +00:00

2.2 KiB
Executable File

tags
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.

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.