| name | granola |
| description | Back up and query Granola meeting transcripts. Refreshes Granola's auto-rotating auth token, downloads new transcripts locally, names them readably, and searches/reads the archive. Use when asked to 'pull granola', 'download my transcripts', 'sync granola', 'back up my meetings', 'search my transcripts', 'find that meeting', or to refresh Granola credentials. |
| version | 1.1.0 |
| user_invocable | true |
Granola
Back up your Granola meeting transcripts to local disk and
query them. Granola encrypts its local storage and uses single-use auth tokens
that rotate on every refresh, so the key trick is: always re-extract a fresh
token from the live app immediately before downloading. The bundled pull.py
pipeline does this for you.
Requires the Granola desktop app installed and logged in (the token is
decrypted from its local keychain entry). macOS only.
Usage
/granola # pull latest (refresh token → download new → rename)
/granola search XYZ # full-text search across transcript bodies
/granola get <id|title|date> # print one transcript
/granola stats # archive location + counts
Trigger phrases: "pull granola", "sync granola", "download my transcripts",
"back up my meetings", "search my transcripts", "find that meeting about …",
"refresh granola token".
Step 0 — Locate or install the toolkit (do this first, every run)
The skill needs the Granola-Download repo (scripts + venv). Find an existing
checkout or install one to a stable location. Run this bootstrap; $GRANOLA
is the repo path for all later steps:
set -e
GRANOLA=""
for d in "$HOME/develop/Granola-Download" "$HOME/Granola-Download" \
"$HOME/.local/share/granola-download"; do
[ -f "$d/pull.py" ] && GRANOLA="$d" && break
done
if [ -z "$GRANOLA" ]; then
GRANOLA="$HOME/.local/share/granola-download"
git clone https://github.com/dennisonbertram/Granola-Download "$GRANOLA"
fi
if [ ! -x "$GRANOLA/.venv/bin/python" ]; then
python3 -m venv "$GRANOLA/.venv"
fi
"$GRANOLA/.venv/bin/python" -m pip install -q -r "$GRANOLA/requirements.txt"
echo "GRANOLA=$GRANOLA"
Use $GRANOLA/.venv/bin/python as the interpreter for every command below.
Step 1 — Pull (default action)
"$GRANOLA/.venv/bin/python" "$GRANOLA/pull.py"
- Re-extracts a fresh token, then downloads only new documents (skips
already-downloaded by
document_id).
- Renames new folders to
YYYY-MM-DD_HHMM_Subject_<id8> (local time, sorts
chronologically). Idempotent.
- Output lands in
$GRANOLA/transcripts_output/.
If a pull fails with a 400 / "Failed to obtain access token": the saved
token went stale (it rotates, single-use). pull.py already re-extracts before
downloading, so just run it again. If extraction itself fails, confirm the
Granola desktop app is installed and logged in.
Step 2 — Search / read the archive
Prefer the granola MCP server tools if present in this session
(mcp__granola__search_transcripts, __list_meetings, __get_transcript,
__sync, __archive_stats). Otherwise use the bundled stdlib reader:
"$GRANOLA/.venv/bin/python" - "$GRANOLA" <<'PY'
import sys; root = sys.argv[1]; sys.path.insert(0, f"{root}/granola")
import archive
out = f"{root}/transcripts_output"
for r in archive.search_transcripts("QUERY", output_dir=out, limit=10):
print(r["date"], "|", r["title"], "| hits", r["match_count"], "\n ", r["snippet"])
PY
Step 3 — Credentials (handle internally, never print)
Refreshing the token just means running pull.py (or extract_config.py, which
rewrites config.json). Never echo config.json, the access token, or the
refresh token into the conversation — they are live credentials. If the user
explicitly wants to see them, confirm first and point them at the gitignored
config.json.
Reference
| File | Purpose |
|---|
pull.py | One-command pipeline: extract token → download → rename |
extract_config.py | Decrypt Granola storage → fresh config.json (token + client_id) |
granola/download_transcripts.py | Transcript-only download (main.py = full export incl. AI notes) |
granola/rename_folders.py | Idempotent readable-name renamer |
granola/archive.py | Stdlib reader: list_meetings / search_transcripts / get_transcript / stats |
mcp_server/server.py | Local stdio MCP server exposing the above |
Guardrails
config.json, transcripts_output/, and .venv/ are gitignored — never
commit them or paste their contents.
- The token rotates constantly; don't cache it. Re-run
pull.py.
- Full export (AI summaries + workspace/folder metadata):
"$GRANOLA/.venv/bin/python" "$GRANOLA/granola/main.py" ./my-export.