johncarpenter/meetings

Meetings — Granola MCP Integration

Query, retrieve, and save Granola meeting notes locally via the proofgeist Granola MCP server. This server reads from Granola's local cache file — no API connection required.

Important: Local Cache Limitations

The proofgeist MCP reads from ~/Library/Application Support/Granola/cache-v3.json:

• All meeting metadata is available (~300+ meetings)
• Only ~15 recent meetings have transcripts cached locally
• Older transcripts are stored in AWS and not accessible via this MCP

Available MCP Tools

The Granola MCP (mcp__granola__*) exposes these tools:

| Tool | Purpose | Parameters | |------|---------|------------| | search_meetings | Search meetings by keyword, participant, or content | query (string), limit (int, default 10) | | get_meeting_details | Get comprehensive meeting metadata | meeting_id (string) | | get_meeting_transcript | Get full transcript with speaker identification | meeting_id (string) | | get_meeting_documents | Get meeting notes, summaries, and structured content | meeting_id (string) | | analyze_meeting_patterns | Analyze patterns across meetings | pattern_type ('topics'/'participants'/'frequency'), date_range (optional) |

Workflows

1. Find a Meeting

Search to locate the target meeting:

mcp__granola__search_meetings(query="quarterly planning", limit=10)
mcp__granola__search_meetings(query="Circuit", limit=5)
mcp__granola__search_meetings(query="John", limit=10)

Present results as a concise list (title, date, attendees). Confirm with the user if ambiguous.

2. Retrieve Meeting Details

With a meeting_id, pull the content:

mcp__granola__get_meeting_details(meeting_id="<uuid>")
mcp__granola__get_meeting_documents(meeting_id="<uuid>")
mcp__granola__get_meeting_transcript(meeting_id="<uuid>")

3. Save Meeting Summary Locally

The key deliverable — copy meeting content to a local .md file.

Markdown format:

# [Meeting Title]

**Date:** YYYY-MM-DD HH:MM
**Attendees:** Name1, Name2, Name3

## Summary
[AI-generated summary from meeting documents]

## Key Decisions
- [Decisions extracted from notes]

## Action Items
- [ ] [Action item with owner if available]

## Notes
[Human-written notes if present]

## Transcript
[Include only if user requests it — transcripts can be large]

Filename convention: YYYY-MM-DD-meeting-title-slug.md

Save locations:

• Client meetings: clients/<ClientName>/meetings/
• Internal meetings: operations/meetings/

4. Analyze Meeting Patterns

Use the pattern analysis tool for insights:

mcp__granola__analyze_meeting_patterns(pattern_type="participants")
mcp__granola__analyze_meeting_patterns(pattern_type="topics", date_range={"start_date": "2026-01-01", "end_date": "2026-01-31"})
mcp__granola__analyze_meeting_patterns(pattern_type="frequency")

5. Batch Export

For multiple meetings:

1. search_meetings to get the set
2. Iterate through each meeting_id
3. Retrieve details and documents for each
4. Save all to appropriate subdirectory

Tips

• Transcript availability: Only ~15 recent meetings have transcripts in local cache. Older meetings will have metadata but no transcript.
• Speaker labels: Transcripts include speaker identification — preserve this in output.
• Documents vs Transcript: Use get_meeting_documents for notes/summaries, get_meeting_transcript for the full conversation.
• Timezone: All timestamps display in local timezone automatically.

Error Handling

• No transcript available: Meeting may be older than the cache window. Note this to the user.
• Empty results: Broaden the search query.
• MCP not connected: Restart Claude Code to reload MCPs. Check .mcp.json config.

Current Configuration

Server installed at: mcp/granola-ai-mcp-server/ Config in: .mcp.json

{
  "granola": {
    "command": "/Users/john/Documents/Workspace/2Lines/knowledge-base/mcp/granola-ai-mcp-server/.venv/bin/granola-mcp-server",
    "args": []
  }
}