Build guide

Build a Mobile Game with Claude Code

TRASH: a Crossy Road-style mobile party game built end to end with Claude Code and Expo.

Claude CodeExpo
Sep 23, 2026
Optimized for

What you'll build

TRASH - a mobile party-race game in the Crossy Road / Fun Run lane. You run down a street, manhole covers drop if you step on them dead-center (they just wiggle if you clip the edge), and the concept is 10-15 friends each pitch in five bucks, race, and whoever makes it to the end splits the pot. I went from a voice memo to a playable beta running on my actual iPhone in a couple hours, with no technical skills whatsoever. The graphics aren't wild - but the game design and the metrics I wanted actually work.

What you'll need

Step 1: Capture the idea the second you have it

I thought of this game while I was sleeping. Instead of letting it die, I took down a five-minute voice memo rambling the entire concept - the pothole mechanic, the physics, collecting cash and coins on the way, coins revive you. The game metrics. That ramble is the whole spec. Don't overthink this part; just get it out of your head verbatim.

Step 2: Generate mockups with GPT Image

Before any code, I went to ChatGPT and asked for a visual target:

Hey, use gpt-image2. Sketch up a couple iOS designs. I'm trying to design
a game that's very similar to Crossy Road... I want it to be called
something like Trash.

I described the features too, but honestly that wasn't even necessary for a mockup. It came back with something that looked sick, so I kept going - "this is really good, can you add a couple of things?" - until I had the game layout, the title screen, and the icon. It genuinely looked like a game people would play off the App Store.

Step 3: Expect a fight on the graphics

This was the hardest part. I assumed that if I handed Claude the GPT mockup, it would replicate the design in the game. I was wrong about that. It took a bunch of reiterations to get it close, and I'm still not 100% happy - but it's way better than where it started. Two things that help:

Step 4: Pull in a game-dev skills repo to fix your physics

My first major problem: the character just got stuck. Couldn't move forward, couldn't move left or right. (The eventual root cause was beautiful - a single hard-coded 0 in the movement code that should have been a 1.)

The fix: I asked Claude to find me a GitHub repo about game design - physics, game feel, UI. It found awesome-gamedev-agent-skills by Abhishek: 73 skills, 1.1K stars. I had Claude scan all the files and keep only the three that mattered - physics tuning, game feel, and game UI/UX - and implement those into the game. After that I could actually move around.

Before implementing anything from a public repo, have your best model run a thorough security check on it - read through every file and confirm it's safe before saving anything down. You don't want to sacrifice security for a shortcut.

Step 5: Set up the stack - Claude Code + Expo Go

Claude Code is the main tool. Expo Go is the other half: a free sandbox for iOS apps that lets you run your build on a real phone.

  1. Sign up for a free account at expo.dev, and log into the same account in the Expo Go app on your phone.
  2. Go to Expo's Docs section and paste the entire docs into Claude Code. It reads through the MCP server, AI agents, Expo skills - all of it - and knows how to integrate. No API key needed.

Step 6: Get it on your phone

The only manual piece is one line in the terminal (Command + Space, type "terminal"). From the folder where the game code lives:

npx expo start

That spits out a QR code. Scan it with your phone's camera - as long as you're on the same Wi-Fi, the game opens in Expo Go and you're playing it. That's the whole deploy.

Step 7: Know what it costs

After the setup it's just tabs and prompting - collecting references, scoping what a V1 actually needs, then letting the agents cook. I was building my personal website in another tab most of the time.

Step 8: (Optional) AI art for the polish

Two extras I layered on at the end:

Step 9: Decide if the App Store is worth it

Pushing to the App Store needs an Apple developer portal, and review has gotten a lot more strenuous because everyone's shipping AI-generated slop right now. This isn't the game I'd publish - but the beta on my phone was the point.

That's the whole loop: idea → voice memo → mockups → Claude Code → your phone, in a couple hours to a weekend. Your V1 is going to be really, really bad, and you'll reiterate a lot - but you get to see what it could look like and feel like, and get feedback right off the rip. The age of overthinking is over. Good luck and happy building.

Resources

← Back to the library Good luck and happy building