How to build a trading bot with Claude Code (without writing code)

Use one master prompt and small verified steps so Claude Code builds a full AI trading desk: dashboard, safe order gateway, agents and journal.

Why Claude Code

Claude Code runs on your server, reads and writes files, runs commands and checks results. You describe the system and approve each step; it writes the code. You stay in charge of the decisions that matter: limits, accounts and when to go live.

Start with a plan, not code

The master prompt asks Claude to show the folder structure and the docker-compose file first and to wait for your OK. Reviewing the plan before any code is written is the single habit that saves the most time.

Build in small, checked steps

Each step ends with a check you can see: a container running, a page loading, a test order refused because it has no stop. If a check fails, paste the error back and ask for a fix. Commit each working step so you can always go back.

Make the safety parts non-negotiable

Ask for an order gateway that refuses orders without a stop, above a maximum size, after the daily loss limit or after a drawdown stop — and that refuses to trade if the account is not a demo. Then ask Claude to prove each refusal with a test.

Keep secrets out of the chat

API keys and passwords live in a .env file readable only by your user. Claude reads them from there; you never paste them into the conversation.