Granola Meeting Transcripts
@scald
npx machina-cli add skill @scald/granola --openclawgranola
Access Granola meeting transcripts, summaries, and notes.
Setup
Granola stores meetings in the cloud. To access them locally:
- Install dependencies:
pip install requests
- Run initial sync:
python ~/path/to/clawdbot/skills/granola/scripts/sync.py ~/granola-meetings
- Set up automatic sync via clawdbot cron:
clawdbot_cron({
action: "add",
job: {
name: "Granola Sync",
description: "Sync Granola meetings to local disk",
schedule: { kind: "cron", expr: "0 */6 * * *", tz: "America/New_York" },
sessionTarget: "isolated",
wakeMode: "now",
payload: {
kind: "agentTurn",
message: "Run the Granola sync: python {skillsDir}/granola/scripts/sync.py ~/granola-meetings",
deliver: false
}
}
})
The sync script reads auth from ~/Library/Application Support/Granola/supabase.json (created when you sign into Granola on macOS).
Data Structure
After sync, each meeting is a folder:
~/granola-meetings/
{meeting-id}/
metadata.json - title, date, attendees
transcript.md - formatted transcript
transcript.json - raw transcript data
document.json - full API response
notes.md - AI summary (if available)
Quick Commands
List recent meetings:
for d in $(ls -t ~/granola-meetings | head -10); do
jq -r '"\(.created_at[0:10]) | \(.title)"' ~/granola-meetings/$d/metadata.json 2>/dev/null
done
Search by title:
grep -l "client name" ~/granola-meetings/*/metadata.json | while read f; do
jq -r '.title' "$f"
done
Search transcript content:
grep -ri "keyword" ~/granola-meetings/*/transcript.md
Meetings on a specific date:
for d in ~/granola-meetings/*/metadata.json; do
if jq -e '.created_at | startswith("2026-01-03")' "$d" > /dev/null 2>&1; then
jq -r '.title' "$d"
fi
done
Notes
- Sync requires the Granola desktop app to be signed in (for auth tokens)
- Tokens expire after ~6 hours; open Granola to refresh them
- macOS only (auth file path is macOS-specific)
- For multi-machine setups, sync on one machine and rsync the folder to others
Overview
Granola stores meetings in the cloud and lets you pull transcripts, AI-generated summaries, and notes locally by syncing. This skill enables offline review, keyword searching, and quick access to meeting metadata. Each meeting becomes a folder with structured files for easy reference.
How This Skill Works
Install Python 3 and the requests package, then run the provided sync script to pull Granola meetings to ~/granola-meetings. Optionally set up a clawdbot cron job to auto-sync. Each meeting folder contains metadata.json, transcript.md, transcript.json, document.json, and notes.md, with tokens refreshed by signing into the Granola desktop app on macOS (auth file at ~/Library/Application Support/Granola/supabase.json).
When to Use It
- You need offline access to recent Granola transcripts and notes while traveling or offline.
- You want to search across transcripts or notes for keywords (e.g., client names, topics).
- You need to review meetings by date or title using the local metadata.json.
- You manage multiple machines and want a single Granola dataset synced to others via rsync.
- You want quick reference to AI-generated summaries (notes.md) alongside transcripts.
Quick Start
- Step 1: pip install requests
- Step 2: Run initial sync: python ~/path/to/clawdbot/skills/granola/scripts/sync.py ~/granola-meetings
- Step 3: Optional: set up automatic sync with clawdbot_cron using the provided payload
Best Practices
- Run the initial sync after installing dependencies to populate ~/granola-meetings.
- Keep tokens fresh: open Granola to refresh credentials when tokens near expiry (~6 hours).
- Ensure you sign in on macOS so the auth file at ~/Library/Application Support/Granola/supabase.json is created.
- Limit multiple concurrent syncs on the same dataset to avoid conflicts; use a single source machine for sync when possible.
- Verify the local folder structure (metadata.json, transcript.md, transcript.json, document.json, notes.md) to ensure data integrity.
Example Use Cases
- PM reviews the last 10 meetings offline to prepare for a stakeholder update.
- Engineer searches transcripts for a keyword across all client calls to surface action items.
- Team uses rsync to propagate the Granola folder to a second laptop for field work.
- Data scientist analyzes raw transcripts (transcript.json) to extract topics and sentiment.
- Assistant filters meetings by date to compile a chronological meeting log.
