Getting started with Claude Code: build your app with an AI pair programmer

Claude Code is Anthropic's AI coding assistant. Unlike a chat window, it works inside your project: it reads your files, writes and edits code, runs commands like installing packages or starting your dev server, and explains what it's doing as it goes. For someone turning an idea into an app, it's the closest thing to having a patient developer sitting next to you.

This is a practical first session: install it, point it at a project, and set it up so it builds the way you want.

What you need

• A Claude plan that includes Claude Code: Pro, Max, Team or Enterprise, or a Claude Console account with API billing. The free plan doesn't include it.
• A computer running a recent macOS, Windows 10 or later, or a mainstream Linux.
• Git and a GitHub account, so every change is saved and reversible.
• Node.js, which most web projects need anyway.

You can use Claude Code in the terminal, in the desktop app, in your browser at claude.ai/code, or inside VS Code and JetBrains editors. The ideas below are the same everywhere; the examples use the terminal.

Install it

The recommended installer on macOS and Linux is:

curl -fsSL https://claude.ai/install.sh | bash

On Windows, in PowerShell:

irm https://claude.ai/install.ps1 | iex

Check it worked with claude --version. Installation options change from time to time, so if anything doesn't match, follow the current setup guide in the Claude Code docs.

Open your project

Make a folder for your app (or open an existing one), go into it in the terminal, and start Claude Code:

mkdir my-app && cd my-app
git init
claude

The first time, it opens your browser to log in. Then you're in a conversation with an assistant that can see and change the files in that folder, and only that folder.

Give it a memory with CLAUDE.md

Run /init. Claude Code looks at your project and writes a CLAUDE.md file: notes it reads at the start of every session, like the commands to run your app and the conventions to follow. Edit it and add your own rules in plain English:

# About this app
A booking app for personal trainers. Stack: Vite + React, Supabase, Stripe.

# Rules
- Explain changes in plain English; I'm not a developer.
- Never put secret keys in code. Read them from .env.
- Every database change is a new Supabase migration.
- Ask before adding a new dependency.

Keep it short. A focused page of rules works better than a long essay. You can open it any time with /memory.

Plan first, then build

For anything bigger than a small fix, ask for a plan before any code is written. Plan mode does exactly that: Claude explores the project and proposes a step-by-step plan without editing files until you approve it. Press Shift+Tab to cycle through the modes until you reach plan mode, or start a single request with /plan.

Read the plan. Push back on anything you don't understand or didn't ask for. It's much cheaper to change a plan than to unpick code.

Stay in control of changes

Depending on your version and settings, Claude Code either asks before editing files and running commands, or works more independently. Shift+Tab switches between modes. While you're learning, choose the mode that asks, and read what it wants to do.

If a change goes wrong, press Esc twice (or type /rewind) to go back to an earlier point and restore the code, the conversation, or both. Checkpoints cover files Claude edited. They can't undo things that happened outside your project, like changes to a live database or a payment provider, which is another reason to build against test accounts first.

Checkpoints are a safety net, not a replacement for Git. Commit whenever something works: "Commit this with a clear message" is a perfectly good prompt.

Connect Supabase and Stripe

Claude Code can talk to other services through MCP servers, which are connectors that let it look things up and take actions in those tools. Supabase and Stripe both publish official ones that you sign in to with your account:

claude mcp add --scope project --transport http supabase "https://mcp.supabase.com/mcp"
claude mcp add --transport http stripe https://mcp.stripe.com/

Then type /mcp inside Claude Code and choose to authenticate each one. Point them at development or test projects while you build. For anything with real customer data, Supabase lets you limit the connection to one project and to read-only access. These commands come from Supabase and Stripe's own docs, so check them there if anything has changed.

Keep your secrets secret

• Put API keys in a .env file listed in .gitignore, not in your prompts. Anything you paste into a conversation is saved in its history on your computer.
• Ask Claude to read keys from environment variables and to confirm no secret ends up in front-end code.
• You can block Claude from reading files such as .env using deny rules in its settings. See "permissions" in the Claude Code docs.

A good first session

1. "Read this project and explain how it's structured in plain English."
2. "Plan the smallest version of [the one screen your app needs]. Use made-up data."
3. Approve the plan, let it build, and run the app.
4. "Commit this."
5. Repeat, one feature at a time.

How you phrase those requests matters more than anything else in this series. Next: how to write prompts that get the result you want.