Can Claude Fable 5 Build a Real iPhone App?

This video shows how to build a full iPhone habit tracker app using Claude's brand new Fable 5 AI model without writing any code manually. I demonstrate setting up the project with Expo, integrating native iOS widgets, and running the app on a real device.

If you want to see how AI can take you from idea to a working React Native app with a native feel, this video is for you. I also cover practical setup tips like using Bun, Xcode, and development builds to handle the latest Expo SDK.

What's inside

Complete habit tracker app with reminders, heat graph, and iOS widget
Using Claude Fable 5 AI model for app generation
Setting up Expo project with Bun and choosing SDK version
Handling Expo Go limitations and creating development builds with Xcode
Integrating custom Claude skills for app icon generation
Connecting to MCP server for advanced AI capabilities
Running and testing the app on physical devices and simulators
Tips for crafting high-quality prompts to get good AI results

The app built by Claude Fable 5 is a full habit tracker featuring straight reminders, a heat graph showing consistency, and a native iOS widget that displays live heat graph data on the home screen. This is a real, running app on my iPhone, created entirely by the AI without me writing any code.

The widget integration is a key highlight because it requires native capabilities beyond what Expo Go supports. This shows the power of combining AI-generated code with native development tools to build complex features.

Using Claude Fable 5 AI model for app generation

Claude Fable 5 is Anthropic’s latest and smartest AI model, a tier above previous versions like Opus. It costs $10 per million input tokens and $50 per million output tokens, so careful prompt design is important to avoid waste.

I use the Claude Code desktop app to interact with Fable 5. Instead of asking it to build the entire app blindly, I provide detailed, high-quality prompts and context to get precise, production-ready code. This approach reduces back-and-forth and token usage.

Setting up Expo project with Bun and choosing SDK version

Before generating the app, I set up the environment by installing Bun, a fast JavaScript runtime that simplifies running commands like creating Expo projects. Bun is installed via a simple terminal command.

When creating the Expo app, I choose between SDK 54 (the stable version on App Store) and SDK 57 (latest beta). Because Expo Go on the App Store only supports SDK 54, I opt to create a development build with SDK 57 to access the newest features and widget support.

Handling Expo Go limitations and creating development builds with Xcode

Expo Go currently supports only SDK 54, so to use SDK 57 features like widgets, I create a development build. This involves compiling the app into a standalone binary using Xcode on Mac.

Xcode is required to run the app on simulators or physical devices. Running on a physical device requires an Apple developer account, but simulators can be used without one. This setup allows testing advanced native features that Expo Go cannot handle.

Integrating custom Claude skills for app icon generation

To enhance the AI’s capabilities, I add custom Claude skills like an app icon generator. This skill is maintained by my team and allows Claude to create a unique icon for the app automatically.

I install the skill by pasting a command in the terminal, giving Claude Fable 5 “superpowers” to generate assets beyond code. This integration improves the app’s polish and saves manual design work.

Connecting to MCP server for advanced AI capabilities

I connect Claude to the MCP server, a tool available to Code with Beto Pro members, which provides access to production-tested code patterns and advanced features like haptics, onboarding flows, and animations.

By linking my profile’s MCP setup to Claude, I enable it to use these resources, improving the quality and realism of the generated app. This is optional but recommended for pro users.

Running and testing the app on physical devices and simulators

After generating the project, I run to launch the development server. Using Expo Go (SDK 54) or a development build (SDK 57), I scan the QR code to open the app on my phone.

I demonstrate testing on a physical iPhone with live updates and light mode support. The app runs smoothly, showing the blank Expo template as a starting point before adding features via AI prompts.

Tips for crafting high-quality prompts to get good AI results

Because Fable 5 is expensive to use, I emphasize the importance of detailed, specific prompts rather than vague requests like “build me a habit tracker.” Providing a clear plan, commands, and context helps Claude generate better code with fewer iterations.

I also recommend manually running setup commands to save tokens instead of letting the AI do everything. This practical approach balances AI automation with developer control.

Resources

Pro membershipPro membership

Access MCP server and exclusive Claude skills to supercharge your AI app development.

CourseReact Native course

Learn how to build and ship React Native apps with practical concepts and lifetime access.

PlatanoLaunch-ready React Native template

See a production app with payments and store assets, perfect for learning and shipping fast.

Let's connect!