Build Your Own Credit Card Benefits Tracker

The tracker exists to answer a single question: 'Am I getting more value than the annual fee, and should I keep or cancel this card before renewal?' Defining that decision in one sentence upfront shapes everything else the app does.

Phase 1: structure and manual logging for one card, tested until correct.

Phase 2: add remaining cards.

Phase 3: points redemptions and unusual value types (e.g. Bilt Cash, its own in-ecosystem currency).

Phase 4: Plaid transaction sync.

Phase 5: alerts/digest.

Each phase gets verified before moving on.

Before building, compile a clear list of each card's benefits (one file per card, ideally). Claude can pull the raw list from the internet, but you apply the judgment about what counts as value to you. Add a 'realistic value' field (defaulting to last year's usage) and mark benefits you'd never use as N/A, so the math reflects reality, not marketing.

Claude Code's default UX isn't great. Stop mid-build and prompt: stop building, act as the top UX designer in the world, and show me an HTML mockup of the user experience. Iterating on the mockup first, then resuming the build against the approved design, is much faster than fixing the UI in code. Screenshotting a product you like and feeding it in speeds this up dramatically.

Plaid picks up statement credits automatically (e.g. Resy dining credit) but can't see wallet-style credits that never hit your card (e.g. Uber Cash), so those stay manual. Ambiguous transactions (refunds that look like credits) go into a review queue you confirm or dismiss, and Claude learns your patterns over time. A classifier trained on a year of authorized-user transactions can tag correctly ~90% of the time.

Plaid setup: you sign up as an individual, and it costs roughly $0.50 per account per month, the only paid piece of the whole build. Plaid requires an HTTPS URL, so use a free tunneling tool to expose your local app via a public URL (the app itself stays local).

Alerts via calendar feed: instead of push notifications, the app updates a calendar subscription in Gmail. Expiring credits show up as reminders right on your calendar.

Keep everything reversible: build in the rule that any logged entry can be undone, so testing doesn't corrupt your data.

Two clocks, not one: track the membership year (when your annual fee renews) separately from each credit's reset window (monthly, quarterly, semi-annual, annual).

Time investment: ~10–20 hours of casual weekend building. Most of the back-and-forth is design and copy polish, not core logic. It pays for itself fast: one member surfaced credits she'd never used (Chase's $10 non-restaurant DoorDash credit, a Bilt hotel credit), and another discovered the Amex Venue Collection (10% back at major arenas + fast-track entry) live during the session.