granola-claude-mcp

cobblehillmachine/granola-claude-mcp

3.4

If you are the rightful owner of granola-claude-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Granola-Claude MCP Server connects Granola.ai meeting data to Claude Desktop, enabling natural language queries across meeting history.

Tools
4
Resources
0
Prompts
0

Granola-Claude MCP Server

Transform your meeting data into AI-powered business intelligence

A custom Model Context Protocol (MCP) server that connects Granola.ai meeting data directly to Claude Desktop, enabling natural language queries across your entire meeting history.

License: MIT Python 3.12+ MCP Compatible

šŸš€ Why This Exists

While Granola-MCP was documented across the web with detailed feature lists and installation guides, the actual repository and PyPI package didn't exist. Rather than wait for an official release, we reverse-engineered Granola's local cache structure and built a comprehensive MCP server from scratch.

The result? Ask Claude questions like:

  • "Show me all client meetings from last month where we discussed budget"
  • "Which team members need follow-up from the Enron project meetings?"
  • "Analyze our meeting patterns with external clients vs internal team meetings"

āœØ Features

šŸ” Natural Language Meeting Intelligence

  • Query 273+ meetings using plain English
  • Search across titles, content, and participant lists
  • Instant access to meeting metadata and calendar integration

šŸ“Š Advanced Analytics

  • Meeting statistics and pattern recognition
  • Attendee analysis and engagement tracking
  • Project timeline reconstruction from meeting history

šŸ”— Complete Integration

  • Google Calendar event details and attendee responses
  • Conference links (Google Meet, Zoom) and participation status
  • Rich metadata including organizers, response status, and sharing settings

šŸ›”ļø Privacy-First Architecture

  • All data processing happens locally - no cloud uploads
  • Direct cache file access - no API dependencies
  • Respects existing Granola permissions and access controls

šŸŽÆ Use Cases

Project Management

"Show meetings about the XXI Media project and identify next steps"
ā†’ Surfaces July 16 Zoom meeting with Pyxl, attendee status, action items

Client Relations

"Which clients had 'needsAction' responses this week?"
ā†’ Identifies follow-up requirements across all client touchpoints

Team Analytics

"Who attends the most external client meetings?"
ā†’ Analyzes workload distribution and development opportunities

Meeting ROI

"Compare meeting outcomes between Google Meet vs Zoom sessions"
ā†’ Platform effectiveness analysis for future planning

šŸ› ļø Installation

Prerequisites

  • macOS with Granola.ai installed and configured
  • Python 3.12+
  • Claude Desktop application
  • Active Granola meeting history

Quick Start

  1. Clone the repository

    git clone https://github.com/[USERNAME]/granola-claude-mcp.git
cd granola-claude-mcp

  2. Make the server executable

    chmod +x granola_mcp_server.py

  3. Test the installation

    python3 test_granola.py

    You should see:

    āœ… Cache loaded successfully!
šŸ“Š Statistics: [your meeting data]
šŸ“… Recent meetings: [your recent meetings]

  4. Configure Claude Desktop

    Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

    {
  "mcpServers": {
    "granola-mcp": {
      "command": "python3",
      "args": ["/path/to/granola-claude-mcp/granola_mcp_server.py"],
      "env": {
        "GRANOLA_CACHE_PATH": "/Users/[username]/Library/Application Support/Granola/cache-v3.json"
      }
    }
  }
}

  5. Restart Claude Desktop

    Quit completely (āŒ˜+Q) and restart Claude Desktop.

  6. Test the integration

    Ask Claude: "Show me my recent Granola meetings"

šŸ”§ Available Tools

get_recent_meetings

Get your most recent meetings with full metadata

Parameters: limit (integer, max 50)
Returns: Meeting list with titles, dates, attendees, calendar integration

search_meetings

Search across meeting titles and content

Parameters: query (string), limit (integer, max 20)
Returns: Matching meetings with relevance scoring

get_meeting_details

Deep dive into specific meeting data

Parameters: meeting_id (string)
Returns: Complete meeting information including notes, attendees, conference details

get_statistics

Analytics across your meeting corpus

Parameters: none
Returns: Meeting counts, transcript availability, word counts, trends

šŸ—ļø Architecture

Data Flow

Granola.ai ā†’ Local Cache ā†’ MCP Server ā†’ Claude Desktop ā†’ AI Intelligence

Cache Structure

Granola uses a sophisticated "double-JSON" encoding:

# Standard JSON file containing JSON string
cache_data = json.loads(data['cache'])
state = cache_data['state']

# Three primary data collections:
documents = state['documents']        # 273+ meeting records
metadata = state['meetingsMetadata']  # 28 relationship objects  
transcripts = state['transcripts']    # 16 available transcripts

MCP Protocol Implementation

  • Full MCP 2025-06-18 protocol compliance
  • Proper initialization and capability negotiation
  • Error handling and graceful degradation
  • Real-time query processing with sub-second response

šŸ› Troubleshooting

"Server Failed" in Claude Desktop

  1. Check the logs: Claude Desktop ā†’ Settings ā†’ Developer ā†’ Open Logs Folder
  2. Verify file paths in your configuration
  3. Test server manually: python3 granola_mcp_server.py

"No meetings found"

  1. Ensure Granola.ai has processed meetings
  2. Verify cache file exists: ls -la ~/Library/Application\ Support/Granola/
  3. Check cache file size (should be several MB for active users)

Python/Permission Issues

  1. Verify Python 3.12+: python3 --version
  2. Make server executable: chmod +x granola_mcp_server.py
  3. Check file permissions on Granola cache directory

MCP Connection Issues

  1. Restart Claude Desktop completely (āŒ˜+Q, then reopen)
  2. Validate JSON syntax in claude_desktop_config.json
  3. Check environment variable path matches your username

šŸ¤ Contributing

We welcome contributions! This project fills a real gap in the AI productivity ecosystem.

Ways to contribute:

  • šŸ› Bug Reports: Issues with specific meeting formats or edge cases
  • āœØ Feature Requests: Additional analytics, export formats, integration ideas
  • šŸ”§ Code Contributions: Performance improvements, additional MCP tools
  • šŸ“š Documentation: Setup guides, use case examples, troubleshooting

Development Setup

git clone https://github.com/[USERNAME]/granola-claude-mcp.git
cd granola-claude-mcp

# Test your changes
python3 test_granola.py

# Run the server manually for debugging
python3 granola_mcp_server.py

šŸ“‹ Roadmap

Phase 2: Advanced Analytics

  • Transcript analysis with action item extraction
  • Sentiment analysis for meeting effectiveness
  • Automated follow-up recommendation engine
  • Cross-meeting topic and decision tracking

Phase 3: Enhanced Integration

  • Real-time cache monitoring for live updates
  • Multi-platform meeting data (beyond Google Calendar)
  • Export capabilities (PDF reports, CSV analytics)
  • Dashboard web interface for meeting intelligence

Phase 4: Enterprise Features

  • Multi-user deployment with access controls
  • CRM integration for complete client lifecycle tracking
  • Advanced reporting and business intelligence
  • API development for third-party integrations

šŸ“„ License

MIT License - see file for details.

šŸ™ Acknowledgments

  • Anthropic for the Model Context Protocol specification
  • Granola.ai for building sophisticated meeting intelligence (even if the MCP integration was MIA)
  • The MCP Community for examples and best practices

šŸ“ž Support

ā­ Star this repo if it helps you extract more value from your meetings!

Built with ā¤ļø by the team at Cobble Hill - AI integration specialists