# Goose Hints for Granola MCP Server
## What This Project Is
This is an MCP (Model Context Protocol) server that provides AI agents access to Granola.ai meeting data.
It exposes 7 tools for searching, retrieving, and filtering meetings via the MCP protocol.
## Quick Commands
- Build: `npm run build`
- Watch: `npm run watch`
- Test run: `node index.js`
## Key Architecture Points
- Uses FastMCP framework for MCP server implementation
- OAuth 2.0 with WorkOS for authentication (refresh token rotation!)
- In-memory caching with 5-minute TTL
- ProseMirror to Markdown conversion for meeting notes
- ESM modules throughout (use .js extensions in imports)
## When Modifying Code
1. Always run `npm run build` after TypeScript changes
2. Test token refresh carefully - rotation means old tokens become invalid
3. API responses may use 'docs' or 'documents' keys - handle both
4. MCP tools must return strings (not objects)
## File Purposes
- server.ts: MCP tools and server startup
- tokenManager.ts: OAuth handling (sensitive - be careful)
- apiClient.ts: REST API calls to Granola
- prosemirrorConverter.ts: Content format conversion
- types.ts: TypeScript interfaces
## Common Patterns
```typescript
// Adding a new tool
server.addTool({
name: 'tool_name',
description: 'What it does',
parameters: z.object({
param: z.string().describe('Description')
}),
execute: async ({ param }) => {
await ensureDataLoaded();
// ... implementation
return 'string result';
}
});
```
## External Dependencies
- Granola app must be installed and logged in (for token extraction)
- WorkOS OAuth (client_01JG3HBXEG0QA83Z7XRPGVPFMB)
- Granola API at https://api.granola.ai
