quant

# Quant > Turn your Claude subscription into a personal life-tracker. Talk to Claude, watch your data appear on your dashboard. **Live:** https://quant-frontend.pages.dev --- ## Why I built this I've been tracking time, calories, and expenses since I turned 18 — it's simply beautiful to be aware of what's happening to your body, money, and time. Two years into a full-time job, the friction got too high. I was falling short at least 50% of the time — not from a lack of care, but from juggling three separate apps every evening: - **Lifesum** for calories - **Wallet by BudgetBakers** for expenses - **Toggl Track** for time Three subscriptions. Three different habits to maintain. Enough friction to make me fall off. I wanted one place. One habit. One app. --- ## How it works The core input is a sentence, not a form. Instead of opening an app and filling out fields, you tell Claude what you did today: > _"Had oatmeal for breakfast, chicken rice bowl for lunch. Bought a book for ₹180, weekly groceries for ₹1200. Worked 9–6, gym 7–8, Netflix after."_ An MCP server handles everything from there — parsing the narration, estimating macros, categorizing expenses, computing time durations — and pushes structured data directly to a Cloudflare D1 database. The React dashboard reflects it instantly on refresh. You can also use voice-to-text (e.g. Wispr Flow) and just speak your day. ### Architecture ``` Claude Desktop / Claude.ai │ MCP over Streamable HTTP │ Authorization: Bearer <api_key> ▼ ┌─────────────────────────────────────┐ │ Cloudflare Worker (Hono) │ │ │ │ POST /mcp ← MCP tools │ api_key auth → resolves userId │ POST /api/auth/sync ← first login │ Clerk JWT auth │ POST /api/query/* ← dashboard │ Clerk JWT auth │ │ │ Routes → Repos → DAL → Drizzle/D1 │ └─────────────┬───────────────────────┘ │ Cloudflare D1 (SQLite) users, api_keys, food_logs, expense_logs, time_logs ▲ │ REST POST /api/query/* ┌─────────────┴───────────────────────┐ │ Cloudflare Pages (React + Vite) │ │ Clerk · TanStack Query · Recharts │ └─────────────────────────────────────┘ ``` ### MCP Tools (9 total) | Tool | What it does | | ------------------------- | --------------------------------------------------------------- | | `log_meal` | Bulk-inserts food items for a meal; LLM estimates macros | | `get_food_summary` | Daily totals + per-meal breakdown for a date range | | `log_expense` | Logs a single expense with category and payment method | | `get_expense_summary` | Totals grouped by day and category | | `log_time` | Logs an activity with start/end datetimes (cross-midnight safe) | | `get_time_summary` | Minutes per bucket + full activity list | | `get_today` | Returns today's date — prevents LLM date hallucination | | `list_buckets` | Returns valid time bucket names | | `list_expense_categories` | Returns valid expense category names | ### Auth - **Dashboard (REST):** Google OAuth via Clerk. JWT validated on the Worker edge using `verifyToken` from `@clerk/backend`. - **MCP (Claude Desktop):** A personal API key auto-generated on first login and shown on the Settings page. --- ## Why MCP instead of a built-in LLM call? My first instinct was to embed an LLM API call inside the app — hit the API, parse the response, save to DB. Then I caught myself: I already pay $20/month for Claude. Why add API costs on top? By building an MCP server, I turned the Claude subscription I already have into the compute layer. Claude does the parsing and reasoning; my server just receives structured tool calls and writes to the database. --- ## Getting started ### Prerequisites - [Cloudflare account](https://dash.cloudflare.com) (free tier works) - [Clerk account](https://clerk.com) (free tier works) - Node.js - Wrangler CLI: `npm i -g wrangler` ### Backend ```bash cd mcp-server npm install # Create D1 database wrangler d1 create quant-db # Paste the database_id into wrangler.jsonc # Set secrets wrangler secret put CLERK_SECRET_KEY # Run migrations npm run db:migrate:local # local dev npm run db:migrate # production # Dev server npm run dev # Deploy npm run deploy ``` ### Frontend ```bash cd frontend npm install # Create a .env.local file: # VITE_CLERK_PUBLISHABLE_KEY=pk_... # VITE_API_URL=https://quant-backend.<your-subdomain>.workers.dev npm run dev # local dev npm run build # production build npm run deploy # deploy to Cloudflare Pages ``` ### Connect to Claude Desktop 1. Sign in at https://quant-frontend.pages.dev — your API key is auto-generated. 2. Go to **Settings** and copy your MCP Server URL and API key. 3. Add to your Claude Desktop config (`claude_desktop_config.json`): ```json { "mcpServers": { "quant": { "url": "https://quant-backend.<your-subdomain>.workers.dev/mcp", "headers": { "Authorization": "Bearer <your-api-key>" } } } } ``` 4. Restart Claude Desktop. The 9 tools will appear automatically. --- ## Database schema ``` users — clerk_user_id (unique), email api_keys — one per user, hex token, revokable food_logs — one row per food item (grouped at query time by date + meal_type) expense_logs — amount, category (int enum), description, payment_method time_logs — bucket (int enum), activity, start_time, end_time (full ISO datetimes) ``` All enum columns store integers in the DB (`.$type<IntEnum>()` in Drizzle) and are converted to human-readable labels at the application layer. --- ## What's next - Task tracking — replacing my Todoist subscription - Habit tracker built around the [12-Week Year](https://12weekyear.com) framework Goal: one dashboard for everything self-development.