granola-ai

Granola MCP Server exposes Granola.ai's meeting intelligence capabilities via the MCP interface, enabling Claude Desktop (and other MCP clients) to query and retrieve rich meeting data such as meeting lists, details, transcripts, notes, and exports from the local Granola cache. The server relies on the Granola CLI tooling (via the uv runtime) to interact with the Granola data locally, offering commands to search, fetch, and export meeting content along with pattern analyses and timezone-aware displays. Use the MCP endpoint through Claude Desktop by configuring your mcpServers section to point to the local uv-based Granola MCP server, then issue MCP requests to retrieve structured meeting information for downstream tooling or workflows.

Prerequisites:

• Python 3.12+ and the uv runtime (uv package manager)
• macOS with Granola.ai installed and a local Granola cache at ~/Library/Application Support/Granola/cache-v3.json
• Claude Desktop (for integration with the MCP server)

Installation steps:

1. Clone the repository and navigate to the project directory:

git clone <repository-url>
cd granola-ai-mcp

2. Install dependencies and verify uv is available (uv is used to run the MCP server):

uv sync

3. Run a quick test to ensure the Granola CLI can interact with the local cache:

uv run granola list

4. Configure Claude Desktop to point to this MCP server by editing claude_desktop_config.json (example shown in README):

{
  "mcpServers": {
    "granola": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/granola-ai-mcp", "run", "granola-mcp-server"],
      "env": {}
    }
  }
}

5. Restart Claude Desktop to load the MCP server and begin querying Granola data via MCP.

Optional global installation steps (for convenience):

• Install the Granola CLI globally via uv (from anywhere):

uv tool install granola-ai-mcp-server

• Or install from local source in editable mode:

uv tool install --editable /path/to/granola-ai-mcp-server

Upgrade steps (when needed):

uv tool install --upgrade granola-ai-mcp-server

Notes and tips:

• The MCP server relies on your local Granola cache; ensure the cache path matches what Granola uses and that the cache file is readable by the MCP runtime.
• Timezone handling: timestamps are displayed in your local timezone when using Granola's CLI tooling via the MCP server.
• If you modify the Granola cache path or environment (e.g., different macOS user), update the claude_desktop_config.json accordingly and restart Claude Desktop.
• Common issues include permissions problems accessing the Granola cache or mismatched directory paths in the uv command arguments. Make sure the absolute path provided in the MCP server command matches your project location.
• Debugging: run uv run granola list (or other commands) directly in the terminal to confirm the MCP server can access Granola data before wiring it into Claude Desktop.