In-depth guide · · 11 min read

Build an app with Claude Code without knowing how to code

Subscription, usage limits, interfaces without a terminal, a brief, five steps and checks: how to have Claude Code build an app when you can't code. Prices and documentation checked on October 10, 2026.

Building an app with Claude Code without knowing how to code is possible, as long as you can describe what you want and check what was done. Claude Code is Anthropic's coding assistant: it reads a project's files, edits them and runs commands based on requests written in plain language. It writes the code. It does not decide for you what the app should do, or whether it does it correctly.

This guide is for someone who will never open a terminal. It covers what comes before the build (subscription, usage limits, interface, brief) and what comes after (checks, blockers, going live). To place Claude Code among the other ways to build an app without coding, start with our general guide.

What you need before you start

$20 per monthClaude Pro, which includes Claude Code
$17 per monthClaude Pro billed annually, i.e. $200 paid up front
from $100 per monthClaude Max, billed monthly

These prices were checked on October 10, 2026 on the Claude plans page, in dollars and excluding tax. The same page states that Claude Code is included in all paid plans. The free plan does not include it. For a first project, the Pro plan is enough to try. It does not guarantee you will finish an app within the month.

The usage limits are more surprising than the price. According to that page, every plan has usage limits that reset on a rolling five-hour window, and paid plans add a weekly limit on top. Claude Code shares these limits with your conversations in Claude: a long discussion in the morning reduces what is left for building in the afternoon. No number of messages or hours is published. So you cannot know in advance how many sessions a subscription will cover. Our article on Claude Code's weekly limits explains what to do when you hit them.

Mac or PC: both work. The Claude desktop app is available for macOS, for Windows (x64 and ARM64 processors) and, in beta, for Ubuntu and Debian, according to the desktop app documentation. It includes Claude Code, with no separate installation, and requires a paid subscription. Also plan a budget for what comes next: hosting the app, a domain name, sometimes an email sending service. These costs are not part of the Claude subscription.

Without a terminal: the interfaces that drive Claude Code

Anthropic offers Claude Code on several surfaces connected to the same engine, listed in the official Claude Code overview. The terminal is one of them. Here are the ones a non-developer can consider, plus ours.

  • The Claude desktop app, Code tab: you choose a folder, write your request, review the changes and watch the app run in a preview pane.
  • The browser, at claude.ai/code, and the Claude mobile app on iOS and Android: the work runs with no local installation, so you can start a long task and come back later.
  • The extensions for the VS Code and Cursor code editors: they show the changes inside the editor, which assumes you are comfortable in front of a list of files.
  • Maestro, which we publish: a working method around the assistant, described below.

For a first project, the desktop app is the simplest starting point. A selector sets how much autonomy the assistant has; the documentation names these modes Manual, Accept edits, Plan and Auto. In Plan mode, Claude prepares its approach before touching any files. In Manual mode, it asks you before each edit or command. The documentation recommends starting a complex task in Plan, then switching to Manual or Accept edits once the plan is approved. For you, that means: read the plan, correct it, and only then let it build.

Maestro is a Mac app (macOS 13 or later), in invitation-only beta and free during that beta. It is not available on Windows at this time. It has Claude Code work within a team of nine agents with distinct roles: an analyst frames the idea, a product manager writes the specification, a developer writes the code and its tests, a quality agent checks each delivery. You approve each step before the next one. AI usage is still counted against your Claude subscription, or billed through the API if you connect a key. If you are torn between two assistants, our article Codex or Claude Code compares their terms.

Describing your app to Claude Code: a sample brief

Claude Code makes better progress with a written description than with a one-line request. Here is a brief for a case invented for this guide: Inès runs ceramics workshops and manages sign-ups through messages and a spreadsheet. Neither she nor her studio exists.

This brief contains four pieces of information Claude Code cannot guess: who uses the app, the business rule (six places at most), what is excluded (payment) and a confidentiality requirement. It says nothing about technology: it is up to the assistant to propose it and explain it to you. The last sentence matters as much as the rest. It asks for questions and a plan, instead of a first draft you would not be able to review.

Some things are deliberately missing from this brief: what happens if Inès cancels an entire workshop? Should participants be notified by email? A good answer from the assistant starts with these questions. If it starts with code, repeat the instruction. Our sample specification shows how to write rules, edge cases and what stays out of the project, without technical vocabulary.

The 5 steps of a first project

To build an app with Claude, split the project into short steps, each one checkable before moving to the next. Here is the order we recommend for a first project, applied to Inès's case.

1. The plan

Send the brief in Plan mode and answer the questions. Read the proposed plan as you would a quote: does each announced screen match a need in the brief? An addition you did not ask for, such as customer accounts or a newsletter, gets removed now, not after it has been built. Ask for the plan to be saved in a project file: it will serve as memory for the following sessions.

2. The data

Before the screens, have it describe what the app will store. A workshop has a date, a time and a number of places; a registration has a name, an email, a phone number and a workshop. Ask where this data will be stored and who will be able to read it. A large part of security is decided here, and it is one of the most expensive parts to rework later.

3. The screens

Have one flow built at a time: first the public sign-up page, then the list of participants. In the desktop app, the preview shows the screen while Claude works. Look at it through a participant's eyes: is the text in the right language, the date readable, the button clear? Write your corrections in full sentences and send them together, rather than one by one.

4. The tests

Prepare a fictional workshop with five places already booked. Register a sixth person, then a seventh: the seventh must be refused. Close the app and reopen it: the registrations must still be there. Also ask Claude Code to write automated tests for these rules. They will be useful at every future change, without replacing your own test.

5. Going live

An app running in the preview only runs on your computer. To make it reachable, you need a host, often a domain name, and an online database. Ask Claude Code to guide you step by step, but create the accounts yourself, in your own name, and never paste a secret key into the code. In one of our testers' projects, we found a key visible in the code of their app. Also keep control of the code and the access: if you can't leave, you don't own it.

Checking what Claude Code built

When Claude Code announces that it is done, it describes what it thinks it did, without proving it. An advertising professional who doesn't code reported getting more than seven hundred passing tests on an app whose import skipped one file in three. We analysed his case in everything is green and yet it's wrong. Tests check what they were asked to check, nothing more.

Three checks are within your reach without reading the code:

  • Compare the numbers: if you import forty workshops from your spreadsheet, the app must show forty.
  • Replay the edge cases from the brief: the full workshop, the cancellation, the phone field left empty.
  • Test access: signed in as a participant, try to open the list of participants. You must not be able to see it.

That last point deserves more than one test. Our guide is your customer data protected offers three simple checks and the evidence to ask of whoever builds the app. You can also ask Claude Code, in a new conversation, to review its own work from a security angle and list what it could not verify. If that list is empty, ask why.

When you hit the limit or the AI goes in circles

Two blockers come sooner or later. The first is the usage limit. Write down the exact message, with the window concerned and the reset time if they are shown. Before paying more, check which account is being used. Anthropic's help page on Claude Code with Pro or Max states that if an ANTHROPIC_API_KEY variable is set on your computer, Claude Code uses that key instead of the subscription, and usage is billed to the API. The usage credits offered when the limit is reached are a paid extra: set a maximum amount before enabling them.

The second blocker is harder to spot: the assistant fixes one problem, creates another, then goes back to the first. Stop the series of requests. Describe what worked before, what no longer works and since which change, then ask for one fix at a time. Our article the AI fixes one thing and breaks another details the method. If the conversation has become very long, a new conversation that starts from the saved plan can help you get back to a clear base.

Claude Code, Lovable or Cursor: which one is for you?

Lovable works in the browser: you describe, a preview appears, hosting is offered on the spot. It is the shortest path to try an idea in one evening. Cursor is a code editor with a built-in agent; it suits someone who wants to learn to follow their project's files. Claude Code works on a folder on your computer, or in the cloud from the browser, with a desktop app that shows the changes and a preview.

Choose the tool in which you understand what is progressing and know how to check the result, even if it does less than the others. Our Lovable vs Claude Code comparison gives the recorded prices and the differences in code and local work. Claude Code or Cursor covers the other choice for a non-developer.

Where to start

Choose a small, useful project whose business you know by heart. Write its brief on one page, with at least one business rule and one excluded item. Open the Claude desktop app in Plan mode, or request access to Maestro if you prefer to be guided step by step on a Mac. Either way, the first session is a success if you can explain what was built and how you checked it.

Back to contents

Read the complete guide: build an application without coding

Back to the journal

Take the baton.

Leave your email to try Maestro in the first waves.

The beta is open by invitation on macOS 13 and later. Leave your email for an upcoming wave of access. Windows is in development.

The beta is currently available on macOS 13 or later. Your answer helps us plan other versions.

Your email is only used to let you know when access opens. Nothing else, we promise.