Granola Meeting Intelligence

Access Joel's Granola meeting notes via the granola CLI, which wraps the Granola MCP server through mcporter.

Setup

Binary: ~/.local/bin/granola
Source: ~/Code/joelhooks/granola-cli
Config: ~/.config/granola-cli/mcporter.json
Transport: MCP via mcporter (NOT direct API, NOT npm package)
Auth: Uses Granola's local credentials automatically

CLI Commands

List meetings

granola meetings [--range this_week|last_week|last_30_days|custom] [--start YYYY-MM-DD] [--end YYYY-MM-DD]

Get meeting details (summary, participants)

granola meeting <meeting-id>

Get full transcript

granola meeting <meeting-id> --transcript

Search meetings

granola search "<query>"

Check auth/connection

granola --help

Returns "connected": true if MCP transport is working.

Output Format

All commands return agent-first JSON with HATEOAS next_actions:

{
  "ok": true,
  "command": "granola meetings",
  "result": { ... },
  "next_actions": [
    { "command": "granola meeting <id>", "description": "Get details" }
  ]
}

Important Notes

Timestamps are UTC. Granola returns all dates in UTC. A meeting showing "5:01 PM" on Feb 26 is actually 9:01 AM PST. Always convert when displaying to Joel or writing to Vault files.
Discover actual access: Use Executor’s saved Granola connection and current integration contract. Historical sandbox EPERM is not a universal capability limit. Start with summaries; raw evidence needs the required scoped drilldown receipt.

Known Limitations

Rate limiting: get_meeting_transcript endpoint rate-limits aggressively. Don't batch transcript pulls — space them out.
Concurrency: Keep concurrency at 1 for transcript fetches. Original concurrency of 3 overwhelmed the API.
Granola app not required locally: The MCP server handles auth via stored creds, Granola desktop app does NOT need to be running.

MCP Tools (underlying)

The granola CLI wraps these MCP tools:

query_granola_meetings — NL search with citations
list_meetings — by time range
get_meetings — details by ID array
get_meeting_transcript — verbatim transcript by ID

Pipeline Integration

When a new meeting is detected, fire an Inngest event:

curl -X POST "http://localhost:8288/e/$EVENT_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "meeting/noted",
    "data": {
      "meeting_id": "<id>",
      "title": "<title>",
      "date": "<YYYY-MM-DD>",
      "source": "granola",
      "participants": ["Person A", "Person B"]
    }
  }'

Inngest functions that process meetings:

check/granola-meetings — event-triggered check for new meetings
granola-check-cron — hourly cron (7 * * * *) polling for new meetings
granola-backfill — bulk backfill all meetings

ADR-0055: Granola Meeting Intelligence Pipeline
Source repo: ~/Code/joelhooks/granola-cli
Inngest functions: packages/system-bus/src/inngest/functions/check-granola.ts
Meeting analysis: packages/system-bus/src/inngest/functions/meeting-analyze.ts
Transcript indexing: packages/system-bus/src/inngest/functions/meeting-transcript-index.ts

granola