How to Build Apps in ChatGPT (And Reach +800m Users)

Hey builders,

ChatGPT's public App Store is about to go live.

We've seen multiple references from the OpenAI team that the launch is imminent. We've even just got a sneak peek of the approval process (I'll share more on this soon).

The last time we had an opportunity like this was the iPhone App Store. Instant access to billions of users. First-mover advantages that shaped entire industries.

The same thing is happening now with ChatGPT.

You can now have your app discovered by the 800m+ (and growing) ChatGPT user base. Zero clicks to discover. Zero friction to use.

We've been building apps with OpenAI’s Apps SDK for the last 6+ weeks, both ready for the launch and for our clients. We've been getting asked a TON on how to do it, so we're releasing everything we've learnt to date.

Today, I'm sharing a full walkthrough showing exactly how to build your first app - whether you're technical or non-technical.

Here's what we're covering:

→ What's actually possible to build
→ How apps in ChatGPT work (MCPs, widgets, iFrames)
→ Good vs bad app design principles
→ Common errors and how to debug them
→ Resources you need

Let's dive in.

Check out the full walk through here:

What's Actually Possible to Build

Before we get into the technical details, I want to show you what's possible.

We've been documenting apps being built in this ecosystem. Everything from proposal generators to habit trackers to lead generation tools.

You can see examples and resources at gptappbuilders.com.

The range is massive. And we're still in the early days.

Major companies have already partnered with ChatGPT to build apps: Shopify, Booking.com, Canva, Zillow, Spotify. Very soon, it opens to everyone.

It’s important to mention, the apps that will exist are both B2C and B2B. Therefore, these apps are relevant to everyone. This could be a new channel for your business or perhaps a suite of apps you can use and build internally to make your company work more efficiently.

How Apps in ChatGPT Actually Work

This is the part most people skip. Don’t…

If you understand the architecture, you can debug anything. If you don't, you'll be stuck every time something breaks.

Here's how it works:

Apps are iFrames served directly in the chat interface.

When you use Canva or Spotify inside ChatGPT, you're seeing their interface embedded right in the conversation. It's not ChatGPT rendering their content. It's literally Canva's own UI living inside the chat window.

MCPs (Model Context Protocol) are how ChatGPT talks to your app.

Think of MCPs like an API that lets natural language models communicate with outside services. When ChatGPT recognises that a user wants to interact with your app, it calls your MCP server.

Your MCP server is essentially a little computer that listens for requests from ChatGPT. When someone says "Show me the time" and you've built a time app, ChatGPT sends that request to your MCP.

Widgets are the UI your app sends back.

When your MCP receives a request, it responds with a widget. A widget is just a piece of user interface specifically designed to display inside ChatGPT.

The complete flow looks like this:

• Step 1: User has intent ("I need a playlist for my party tonight")

• Step 2: ChatGPT recognises the intent and calls your MCP server

• Step 3: Your MCP server loads the appropriate widget

• Step 4: That widget component gets loaded into the iFrame in ChatGPT

• Step 5: User sees your app's interface directly in the conversation

That's it. User intent → MCP call → Widget response → Display in chat.

The beautiful thing is once you understand this flow, you can build anything. A proposal generator follows the same pattern as a pizza ordering app. The architecture is identical. Only the widget changes.

Good Apps vs Bad Apps

Not everything belongs in a chat interface.

OpenAI has been very clear about what works and what doesn't. And if you backwards engineer their guidelines, you get a framework for deciding what to build.

The key principle: your tool should feel like a natural extension of ChatGPT, not a bolt-on experience.

Common Errors and How to Debug Them

You will run into errors. Here are the most common ones:

Widget not loading: This usually means the widget server URL isn't configured correctly. The UI is trying to load from localhost instead of your public ngrok URL. Set the base URL to your ngrok URL in the start command.

MCP not connecting: Make sure your ngrok tunnel is still running. Free ngrok URLs expire. You may need to restart ngrok and update the connector URL in ChatGPT.

App not appearing when prompted: Try explicitly calling your app ("using the [app name] MCP"). In dev mode, ChatGPT sometimes needs the explicit instruction before it learns to recognise your app automatically.

General debugging tip: If you’re just getting use to Cursor, use "ask" mode instead of "agent" mode when debugging. Agent mode will aggressively make code changes. Ask mode will help you understand what's wrong without changing everything.

Full Build Walkthrough

We recorded a complete video showing exactly how to build your first app from scratch.

It covers everything above in detail - from the first prompt to seeing your app live in ChatGPT.

Even if you're completely non-technical, you can follow along and get a working app by the end.

Resources

→ Official Apps SDK Documentation - The source of truth from OpenAI

→ OpenAI's Official GitHub Examples - Working examples to feed to your coding agent

→ gptappbuilders.com - Examples and resources for app builders

→ Cursor - The coding IDE with built-in AI agents

→ Node.js - Required runtime for building apps

→ ngrok - Creates a public URL for your local app

This window of opportunity won't last for long.

Start simple. Build something that works. Then layer complexity.

Can't wait to see what you build.

Feel free to drop me a reply with feedback or questions. I read every reply!

Happy building,

Elliot

Learn how AI Agents & Applications can grow your business.

If your company or brand is interested in getting an app inside ChatGPT, we'd love to talk to you. We have limited spots available to work with ambitious teams who want to win distribution inside of this new platform.

Book a strategy call with our team →