granola-meeting-notes

分类: 数据与AI | 上传者: cathrynlavery | 下载: 0 | 版本: v1.0（最新）

Use when the user asks about meeting notes, transcripts, action items, project lists from meetings, meeting prep, weekly digests, or mentions Granola. Also use when asked "what did we discuss", "who was in the meeting", "what did I commit to", "what were the decisions", "recap my meetings", "who said what", or wants to search/export meeting history. Triggers on keywords like meeting, discussed, action items, transcript, recap, digest, attendees, commitments.

更新日志: Source: GitHub https://github.com/cathrynlavery/granola-skill

目录结构

当前层级: 根目录

📁 bin/
- 📄 granola.js 11.4 KB
📄 install.sh 1.5 KB
📄 README.md 4.7 KB
📄 SKILL.md 7.1 KB

SKILL.md

---
name: granola-meeting-notes
description: Use when the user asks about meeting notes, transcripts, action items, project lists from meetings, meeting prep, weekly digests, or mentions Granola. Also use when asked "what did we discuss", "who was in the meeting", "what did I commit to", "what were the decisions", "recap my meetings", "who said what", or wants to search/export meeting history. Triggers on keywords like meeting, discussed, action items, transcript, recap, digest, attendees, commitments.
---

# Granola Meeting Notes

Access Granola meeting notes, transcripts, and summaries from Claude Code.

**Primary:** Granola MCP tools (`mcp__granola__*`)
**Fallback:** Local CLI at `./bin/granola.js` (relative to this skill's directory)

---

## MCP Tools (Primary)

Granola has an official MCP server at `https://mcp.granola.ai/mcp`. When connected, tools appear as `mcp__granola__*` in your session.

### Setup

```bash
claude mcp add granola --transport http https://mcp.granola.ai/mcp
```

You'll be prompted to authenticate via Granola on first use.

### Available Tools

| Tool | Purpose | Key Parameters |
|------|---------|----------------|
| `mcp__granola__list_meetings` | List recent meetings with title, date, attendees | `days` (number) |
| `mcp__granola__get_meetings` | Search meetings by content, notes, transcripts | search terms |
| `mcp__granola__get_meeting_transcript` | Fetch full raw transcript for a meeting | meeting `id` |
| `mcp__granola__query_granola_meetings` | Natural language queries across all meetings | `query` (string) |

### Usage Pattern

```
1. mcp__granola__list_meetings({ days: 7 })            → browse recent meetings
2. mcp__granola__get_meeting_transcript({ id: "..." })  → get full transcript
3. mcp__granola__query_granola_meetings({ query: "..." }) → semantic search
```

---

## Prompt Recipes

Step-by-step patterns using MCP tools as the primary method.

### "What did we discuss yesterday?"

1. `mcp__granola__list_meetings({ days: 2 })` → find yesterday's meetings
2. `mcp__granola__get_meeting_transcript({ id })` → get full transcript for each
3. Summarize: key topics, decisions made, action items

### "Give me my action items from this week"

1. `mcp__granola__list_meetings({ days: 7 })` → get all meetings this week
2. `mcp__granola__get_meeting_transcript({ id })` for each meeting
3. Scan for commitments — look for: "I'll do", "can you", "let's", "by Friday", "action item", "follow up", "need to"
4. Group by owner and present as a checklist with meeting source

### "Prep me for my meeting with [person]"

1. `mcp__granola__query_granola_meetings({ query: "[person name]" })` → find meetings with that person
2. Get transcripts for the most recent 2-3 matches
3. Summarize: last meeting date, what was discussed, open action items, unresolved topics

### "Weekly digest"

1. `mcp__granola__list_meetings({ days: 7 })` → get all meetings
2. Fetch transcripts for each meeting
3. Compile: meetings by day, key decisions, action items by person, recurring themes

### "Compile a project list from my meetings"

1. `mcp__granola__list_meetings({ days: N })` (user specifies timeframe, default 14)
2. Fetch transcripts for each meeting
3. Produce structured markdown:
   - Individual meeting summaries (title, date, attendees, key points, decisions, action items)
   - Master project list (project name, description, status, key people, open items, related meetings)
   - Open action items grouped by person
4. Save to user's specified location (default: `~/Desktop/project-list.md`)

### "What decisions did we make about [topic]?"

1. `mcp__granola__query_granola_meetings({ query: "[topic]" })` → find relevant meetings
2. `mcp__granola__get_meeting_transcript({ id })` for each match
3. Extract decision segments — look for: "let's go with", "we decided", "agreed", "the plan is"
4. Present chronologically with meeting date and context

### "Who said what about [topic]?"

1. `mcp__granola__query_granola_meetings({ query: "[topic]" })` or list + fetch transcripts
2. Filter for segments mentioning the topic
3. Attribute by speaker: `source: "microphone"` = the user, `source: "system"` = other speakers
4. Present as attributed quotes with timestamps

### "Compare meeting A vs meeting B"

1. Get both transcripts via `mcp__granola__get_meeting_transcript`
2. Summarize each independently
3. Highlight: what changed, new decisions, shifted priorities, contradictions

### "Export meetings for [person/project]"

1. `mcp__granola__query_granola_meetings({ query: "[person/project]" })` → find relevant meetings
2. Fetch transcripts and compile into markdown
3. Save to user-specified location or `~/Desktop/meeting-exports/`

---

## CLI Fallback

Use the CLI only if MCP tools are not available in the session (e.g., server disconnected or auth expired).

### Prerequisites

- Granola desktop app installed and logged in on macOS
- Node.js 18+

### ESM Fix

If you get ESM/import errors, copy the CLI to `/tmp` first:

```bash
cp ./bin/granola.js /tmp/granola.cjs
```

Then use `node /tmp/granola.cjs <command>` for all commands below.

### CLI Reference

**Location:** `./bin/granola.js` (relative to this skill's directory)

| Command | Description |
|---------|-------------|
| `list [--days N]` | List meetings from last N days (default: 7) |
| `show <id> [--transcript]` | Show meeting details; `--transcript` fetches full transcript |
| `search <query>` | Search by title, notes, or attendee name |
| `export <id> [--output DIR] [--transcript]` | Export meeting to markdown file |
| `dump [--days N] [--output FILE]` | Bulk export all meetings + transcripts to JSON |

Partial IDs work (first 8 characters).

### CLI Examples

```bash
node /tmp/granola.cjs list --days 7
node /tmp/granola.cjs show abc12345 --transcript
node /tmp/granola.cjs search "Sarah"
node /tmp/granola.cjs dump --days 14 --output /tmp/meetings.json
```

---

## Architecture

| Component | Source |
|-----------|--------|
| All meeting data (MCP) | Granola cloud API via `https://mcp.granola.ai/mcp` (HTTP transport) |
| Auth (MCP) | Handled by Granola's MCP server (browser auth on first connect) |
| Metadata (CLI fallback) | Local cache: `~/Library/Application Support/Granola/cache-v*.json` |
| Transcripts (CLI fallback) | API: `POST https://api.granola.ai/v1/get-document-transcript` |
| Auth (CLI fallback) | WorkOS token: `~/Library/Application Support/Granola/supabase.json` |

Transcript segments have `text`, `source` ("microphone" = user, "system" = others), `start_timestamp`, `end_timestamp`. Transcripts are speech-to-text — names and technical terms may be misspelled.

---

## Troubleshooting

| Problem | Fix |
|---------|-----|
| MCP tools not in session | Run `claude mcp list` — if granola not connected, re-add with `claude mcp add` |
| MCP auth expired | Remove and re-add: `claude mcp remove granola && claude mcp add granola --transport http https://mcp.granola.ai/mcp` |
| ESM/import errors (CLI) | `cp` CLI to `/tmp/granola.cjs` and run from there |
| API returns 401 | Auth token expired — re-open Granola desktop app to refresh |
| Empty transcript array | Meeting wasn't recorded or was too short |
| Missing attendees | Attendee data comes from Google Calendar — check calendar event |

登录后下载/点赞/收藏 ❤ 2 | ★ 0

granola-meeting-notes

目录结构

SKILL.md

举报内容

提示