Shipping a Real AI Product Into the ChatGPT Store
A live travel-risk app and public API — end-to-end, from data pipeline to a published ChatGPT app.
The problem
Why this needed to exist
Anyone planning a trip somewhere uncertain has to wade through government advisory portals, health notices, and stability reports — each in its own format, none of it tailored to who's actually traveling. A solo backpacker, a family, and a business traveler face very different risks from the same advisory, and nobody packages that for them.
I wanted to prove out something bigger too: that I could take a messy public-data problem and ship it all the way to a polished AI product people can actually use inside ChatGPT.
The approach
How I built it
I built the backend on Next.js and Vercel, then exposed it two ways: a public REST API and an MCP server running inside a route handler so ChatGPT can call it directly as an app. The data layer composites several public sources into a single severity score, and a traveler-impact step reframes that score for the specific kind of trip someone is taking.
On top of the product sits the unglamorous-but-essential plumbing of a real SaaS: API-key auth, rate limiting, and tiered freemium access — the difference between a demo and something you can charge for.
- Next.js
- Vercel
- MCP server (ChatGPT app)
- Composite data scoring
- API auth + rate limiting + tiers
The outcome
What it actually does
The app is live in the ChatGPT store, and the same engine powers a public API spanning 13 endpoints and 195+ countries — covering advisories, country comparisons, vaccination lookups, demographics, stability, packing lists, and more.
It's not a prototype gathering dust: the API serves real traffic on a tiered plan, including paying customers building on top of it.
- Published and live in the ChatGPT store
- 13 public API endpoints covering 195+ countries
- MCP server shipped inside a Next.js route handler
- Tiered freemium API with real, paying usage
What I learned
The curve I already climbed
Two things here are genuinely non-obvious until you've done them: standing up an MCP server inside a Next.js app and getting it through ChatGPT-store publishing, and turning a pile of disagreeing public data sources into a single score you can actually defend. Both have sharp edges that cost days to discover the first time.
I've now shipped that full path — data pipeline, public API, auth and billing tiers, and a published ChatGPT app. For a business that wants an AI product (or an MCP integration) rather than a science project, that's the difference between starting from a working map and starting from a blank page.
If you want a real, shippable AI product — API, billing, and a ChatGPT app included — I've already taken one the whole way and know where the path bends.
Want this kind of thing built for your business?
Book a quick call and I'll show you what would actually move the needle for you.