Granola

Setup Granola Integration Skill

Copy this to .agents/skills/setup-granola/SKILL.md for canonical runtime skills. If needed for Claude discovery, the repo bridge .claude/skills -> ../.agents/skills will expose it.

---

---
name: setup-granola
description: Install and configure Granola MCP server for meeting sync. Use when user says "set up Granola", "install Granola integration", or "connect Granola".
---

# Setup Granola Integration

Automatically install and configure the Granola MCP server for syncing meetings to Personal OS.

## Prerequisites Check

Before starting, verify:
1. User is on macOS (Granola is macOS only)
2. User has Granola.ai installed
3. Python 3.12+ is available
4. uv is installed (or can be installed)

## Instructions

### Step 1: Check Prerequisites

Run these checks and report any issues:

```bash
# Check OS
uname -s  # Should be Darwin

# Check Python version
python3 --version  # Should be 3.12+

# Check uv
uv --version  # Install if missing: brew install uv

# Check if Granola cache exists
ls ~/Library/Application\ Support/Granola/cache-v3.json

If uv is missing, offer to install it:

brew install uv

If Granola cache doesn't exist, inform user they need to:

1. Install Granola from https://granola.ai
2. Attend or record at least one meeting

Step 2: Clone MCP Server

Ask user where to clone (suggest ~/Projects):

cd ~/Projects  # or user's preferred location
git clone https://github.com/proofgeist/granola-ai-mcp-server.git
cd granola-ai-mcp-server
uv sync

Step 3: Test Server

Verify installation works:

cd ~/Projects/granola-ai-mcp-server
uv run python test_server.py

Step 4: Create Transcripts Folder

mkdir -p Knowledge/transcripts

Step 5: Install Skills

Copy the Granola skills to the canonical skills folder:

mkdir -p .agents/skills/granola-meeting-sync
cp System/integrations/granola/skills/meeting-sync/SKILL.md .agents/skills/granola-meeting-sync/SKILL.md

Step 6: Update .mcp.json

Read the current .mcp.json file and add the granola server config.

The config should be:

{
  "mcpServers": {
    "granola": {
      "command": "uv",
      "args": [
        "--directory",
        "<GRANOLA_MCP_PATH>",
        "run",
        "granola-mcp-server"
      ],
      "env": {
        "KNOWLEDGE_PATH": "<PERSONAL_OS_PATH>/Knowledge"
      }
    }
  }
}

Replace:

• <GRANOLA_MCP_PATH> with absolute path to granola-ai-mcp-server
• <PERSONAL_OS_PATH> with absolute path to personal-os workspace

Step 7: Verify Setup

Ask user to restart their agent session, then test:

Search my meetings for "test"

or

Check for new Granola meetings

Success Criteria

• Prerequisites verified
• granola-ai-mcp-server cloned and dependencies installed
• test_server.py passes
• Knowledge/transcripts folder exists
• meeting-sync skill installed to .agents/skills/
• .mcp.json updated with granola config
• Agent session restarted
• Granola tools accessible (search_meetings works)

Troubleshooting

Python version too old

brew install [email protected]

uv sync fails

pip install uv  # Alternative installation

Permission denied on cache

Granola must be running with proper permissions. User should open Granola app once.

MCP not loading

Check .mcp.json syntax and ensure paths are absolute (not relative).

After Setup

Inform user they can now:

• Say "Sync my Granola meetings" to sync new meetings
• Daily planning will automatically check for new meetings
• Search across all meeting content