# Granola Import (research, not implemented) Goal: bring older Granola meetings into Hex's meeting format so `hex mcp` can search them beside Hex recordings. Researched from Granola's public docs on 2026-10-15; nothing was logged into, called, or decrypted. ## Official Paths ```ts Granola desktop ├── CSV export // Settings > Profile > Generate CSV │ ├── Emailed link, valid 24 h, opened while signed in; 1 export / 24 h │ ├── Title, summary, transcript, basic details; full history, including │ │ notes older than 30 days on Basic (free) │ ├── Only notes you own that have a summary; no deleted notes │ └── On by default for Basic and Business; Enterprise admins may disable ├── Personal API // https://public-api.granola.ai/v1 │ ├── Bearer grn_ key from Settings > Connectors > API keys │ ├── Business plan members (Enterprise: admin-gated scopes) │ ├── GET /v1/notes (created_after, cursor), GET /v1/notes/{not_…} │ └── GET /v1/notes/{id}/transcript: items with speaker.source │ microphone|speaker, attribution, optional name, start_time/end_time └── Granola MCP // https://mcp.granola.ai/mcp, OAuth ├── Basic: personal notes from the last 30 days; transcript tool paid only └── Business: full history, get_meeting_transcript ``` Sources: [Exporting historical notes](https://docs.granola.ai/help-center/sharing/exporting-notes), [Granola API](https://docs.granola.ai/introduction), [Get Transcript](https://docs.granola.ai/api-reference/get-transcript), [Granola MCP](https://docs.granola.ai/help-center/sharing/integrations/mcp). Granola also documents transcript auto-deletion; check that setting before assuming old transcripts still exist. The local cache on the Pro (`cache-v6.json.enc`, `granola.db`) is encrypted. Do not reverse-engineer it; third-party scripts that do are unofficial. ## Mapping To Hex Granola's macOS transcript sources map directly: `microphone` -> `You` (`MeetingSource::Microphone`), `speaker` -> `Them` (`System`). Absolute `start_time`/`end_time` become offsets from the first item. A note becomes a folder `meetings/granola-/` with a `manifest.json` (status `complete`, `created_at_ms` from the meeting start, `app: "Granola import"`, title), a `transcript.ndjson`, and a `transcript.md` so the folder reads as published. Granola summaries are not Hex notes; keep them in a separate field or a Markdown file rather than impersonating Apple Intelligence `notes`. ## Recommendation 1. Kit runs the CSV export once (free on his plan, full history, no API key). 2. A one-shot importer (`hex meeting import-granola `, debug first) writes folders atomically, skips ids that already exist, and never overwrites. CSV transcripts may lack timestamps or speaker sources; check one real row before designing the parser, and fall back to a single `Them` stream with zero offsets if they do. 3. Use the Personal API only if the CSV lacks speaker sources or times and Kit is on Business; it needs a stored `grn_` key, so ask Kit first. Open questions: Kit's Granola plan, whether the CSV transcript column keeps microphone/speaker attribution and timing, and whether auto-deletion removed older transcripts.