Granola MCP MCP Server

granola-mcp

MCP server for semantic search across Granola meeting notes. Extracts insights, themes (pain-points, feature-requests, decisions, etc.), and key quotes with speaker attribution. Uses LanceDB for fast local vector search.

Based on reverse engineering research by Joseph Thacker and getprobo.

Features

• Export: Extract all your Granola meetings with transcripts
• Semantic Search: Vector-indexed search across meetings with pre-extracted insights
• Speaker Attribution: Distinguishes between host (me) and participants
• Theme Extraction: Auto-categorizes content into themes (pain-points, feature-requests, etc.)
• MCP Server: Exposes search to Claude Code, Claude Desktop, and other AI tools

Prerequisites

• Node.js 18+
• Granola desktop app installed and logged in
• OpenAI API key (for embeddings and insight extraction)

Installation

npm install
npm run build

Quick Start

# One command to sync everything (export + index)
OPENAI_API_KEY=sk-... node dist/index.js sync

# Or with a custom data directory
OPENAI_API_KEY=sk-... node dist/index.js sync ./my-data

# Test search
OPENAI_API_KEY=sk-... node dist/index.js search "user pain points"

CLI Commands

Sync (Recommended)

The easiest way to keep your data up to date - exports from Granola and rebuilds the index in one step:

OPENAI_API_KEY=sk-... node dist/index.js sync

# With options
OPENAI_API_KEY=sk-... node dist/index.js sync ./my-data
OPENAI_API_KEY=sk-... node dist/index.js sync --skip-extraction  # Faster, reuses existing insights

Export from Granola

Export only (without indexing):

node dist/index.js export ./output
node dist/index.js export ./output --format markdown
node dist/index.js export ./output --format json

Build Search Index

# Full indexing with insight extraction (~$0.02/document)
OPENAI_API_KEY=sk-... node dist/index.js index ./export

# Skip extraction (use existing insights, just rebuild embeddings)
OPENAI_API_KEY=sk-... node dist/index.js index ./export --skip-extraction

Search from CLI

OPENAI_API_KEY=sk-... node dist/index.js search "pricing concerns"
OPENAI_API_KEY=sk-... node dist/index.js search "feature requests" --folder "User interviews"

Export for ChatGPT

OPENAI_API_KEY=sk-... node dist/index.js export-combined ./chatgpt.md
OPENAI_API_KEY=sk-... node dist/index.js export-combined ./chatgpt.md --query "user feedback"

Other Commands

node dist/index.js list              # List documents
node dist/index.js workspaces        # List workspaces
node dist/index.js folders           # List folders
node dist/index.js transcript <id>   # Get specific transcript

MCP Server Setup

Claude Code

Add to .mcp.json in your project:

{
  "mcpServers": {
    "granola": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/granola-mcp/dist/mcp/server.js"],
      "env": {
        "GRANOLA_DATA_DIR": "/path/to/granola-mcp/export",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Claude Desktop

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

{
  "mcpServers": {
    "granola": {
      "command": "node",
      "args": ["/path/to/granola-mcp/dist/mcp/server.js"],
      "env": {
        "GRANOLA_DATA_DIR": "/path/to/granola-mcp/export",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

MCP Tools

Speaker Attribution

The system distinguishes between speakers:

• speaker: "me" - The meeting host (you)
• speaker: "participant" - Other people in the meeting

This helps AI understand what's your own idea vs external feedback.

Pre-defined Themes

• pain-points: User frustrations, problems, complaints
• feature-requests: Desired features, wishlist items
• positive-feedback: What users liked, praised
• pricing: Cost concerns, value perception
• competition: Competitor mentions, alternatives
• workflow: How users currently do things
• decisions: Key decisions made, action items
• questions: Open questions needing clarification

Output Structure

export/
├── vectors.lance/           # LanceDB vector index
├── Meeting_Title_1/
│   ├── document.json        # Raw document data
│   ├── notes.md