How to build your first app with Claude Code, step by step
A practical path from an empty folder to a live app, for founders who are not full-time developers
You have an idea for an app, you have heard that Claude Code can write the code, and you are not sure where to start. Most first attempts fail for the same reason: the founder asks for the whole app in one message, gets a pile of code nobody understands, and gives up when the first bug appears. This guide takes a slower route that works: small steps, each one checked before the next.
What Claude Code is
Claude Code is Anthropic's coding agent. You talk to it in plain language, and it reads the files in your project, proposes changes, edits code and runs commands such as tests or a local server. It works in your terminal, and Anthropic also offers it in a desktop app, in the browser and inside editors such as VS Code and JetBrains IDEs.
The important word is agent. Claude Code does not just answer questions; it acts on your files. That is what makes it useful, and it is why you stay in charge of every change.
What you need before you start
- An account that includes Claude Code. According to Anthropic's documentation, that means a Claude Pro, Max, Team or Enterprise plan, or a Claude Console account with pre-paid API credits. The free Claude plan does not include Claude Code.
- A computer. A recent version of macOS, Windows 10 or 11, or a common Linux distribution, with at least 4 GB of RAM and an internet connection.
- A terminal. Terminal on macOS, PowerShell on Windows. If you have never opened one, Anthropic's documentation includes a short terminal guide.
- Git. Git records every version of your code so you can undo mistakes. Install it from git-scm.com. On Windows, Anthropic also recommends Git for Windows so Claude Code can use its Bash tool.
- A free GitHub, GitLab or Bitbucket account to keep your code online and connect it to a host later.
Step 1: install Claude Code
Anthropic's recommended method is the native installer, which also keeps itself up to date. Open a terminal and run the command for your system, as listed in the official setup guide:
- macOS, Linux or WSL:
curl -fsSL https://claude.ai/install.sh | bash - Windows PowerShell:
irm https://claude.ai/install.ps1 | iex
Homebrew (brew install --cask claude-code) and WinGet (winget install Anthropic.ClaudeCode) also work, but they do not update automatically. When the installer finishes, open a new terminal window and run claude --version. If you see a version number, you are ready. Commands change over time, so check the setup page if anything here does not match what you see.
Step 2: start in a project folder
Create an empty folder for your app, move into it and start Claude Code:
mkdir my-first-app, then cd my-first-app, then claude
On the first run it asks you to log in through your browser. Always start Claude Code inside the project folder. It works on the files in the folder where you launch it, so starting from your home folder gives it far more reach than it needs.
Step 3: write a one-page plan
Before any code, write a plan in a file called PLAN.md. Keep it to one page:
- What the app does, in two sentences.
- Who uses it, and the three or four things they must be able to do.
- What is out of scope for the first version.
- The stack, if you have a preference (for example Next.js with a hosted database), or ask Claude Code to propose one and explain why.
Then ask Claude Code to read the plan and point out the gaps. Fixing a vague plan is far cheaper than fixing vague code. A good request: "Read PLAN.md. List the questions you need answered before building, then propose the smallest useful first version." You can also press Shift+Tab, which cycles through the permission modes, to reach plan mode, where Claude Code thinks and proposes without editing files.
Step 4: add a CLAUDE.md
Claude Code starts every session with a fresh context. A CLAUDE.md file in the project root is read at the start of each session and holds what it should always know: the stack, the commands to run, the folder layout and your rules. Run /init inside Claude Code and it drafts one from your project; then edit it. Anthropic suggests keeping each CLAUDE.md under about 200 lines. Our guide to the rules file shows what to put in it, with an example.
Step 5: build one feature at a time
Pick the first feature from your plan and ask for that alone: "Build the sign-up page from PLAN.md. Nothing else yet." Small requests produce small changes that you can actually check. If a session gets long and confused, type /clear and start fresh; CLAUDE.md and PLAN.md carry the important context into the new session.
Step 6: review every diff
Depending on its permission mode, Claude Code asks before it edits a file and shows you the change, called a diff: the lines removed and the lines added. Read it, even if you are not a developer. Ask yourself: is it touching the files I expected? Did it add a package or service I did not ask for? Is there a password or key written into the code? When something is unclear, ask: "Explain this change in plain words." You do not need to understand every line, but you should understand what each change does.
Step 7: run and test
Ask Claude Code how to run the app on your computer, then open it in your browser and use it as a customer would. Try the wrong things too: empty forms, wrong passwords, very long text. Ask it to add automated tests for the important flows and to run them after each change, so a new feature does not quietly break an old one.
Step 8: commit with Git
Each time a feature works, save that version with Git. You can ask Claude Code directly: "Commit my changes with a clear message." A commit is a checkpoint; if the next feature goes wrong, you can return to the last version that worked. Then push the project to a private repository on GitHub or a similar service so it is backed up. Never commit secret files such as .env, and ask Claude Code to confirm they are listed in .gitignore.
Step 9: deploy
Most modern hosts can deploy straight from your Git repository. With Vercel, for example, you import the repository once; after that, every push to a branch gets a preview deployment, and merges into the production branch go live. Netlify, Cloudflare and others work in a similar way. Put your secret keys in the host's environment variable settings, never in the code. Before you share the link, read our guide to common security mistakes in AI-written code and check your app against it.
Skills that help along the way
Skills are packaged instructions that Claude Code loads when a task needs them. A few in the AI Khazna catalogue fit this workflow: Writing Plans and Executing Plans for turning a plan into ordered steps and working through them, Code Review for checking changes against your standards, and Git Guardrails Claude Code, which blocks destructive Git commands before they run. Read each skill before you install it, as you would any code, and browse more in the skills catalogue.
Your next step
Create the folder, install Claude Code and write your one-page plan today, then build only the first feature. When you want more structure, the Vibe Coding hub collects the guides, tools and checklists for taking an AI-built app from idea to launch.