Malou Dongelmans

Episode 01

Claude Code + Cursor setup

First post in “Tools I build with.” Every other episode in this series runs on the foundation below. After a few episodes covering the setup, the series walks through the steps to build a print-on-demand business with AI, without leaning on the usual website builders or marketplace platforms. It might sound contrarian, but really it’s an invitation to become autonomous and creative with AI, and to explore different business ventures at reduced cost.

What Claude Code is

Claude Code is a tool from Anthropic that lets you ask an AI to do things on your computer. You type what you want, in plain English. “Add a contact form to my website.” “Help me understand why my page is loading slowly.” “Build me a page that lists my favourite recipes with photos.” It reads the files you point it at, makes the changes you asked for, and tells you what it did.

Claude Code is not the same as ChatGPT or Claude.ai, even though it comes from the same company. ChatGPT chats. Claude Code works inside a project folder on your computer and actually makes changes, rather than just describing them.

Claude Code can work on files on your own laptop, or on a remote server (a computer you rent, also known as a VPS). The rest of this series covers both. The next post, episode 02, walks through what a VPS is, how to rent one, and why every site on maloudongelmans.com sits on a single rented server. If the word “server” feels intimidating, you’re the exact reader this series is written for.

The site you are reading right now, the journal you are on, every small piece of it: all built using Claude Code. When a post says “I asked Claude to…”, this is what’s being described.

Two ways to run it

Claude Code comes in two flavours. Pick the one that fits how you like to work.

  • Claude Code CLI. A version that runs inside a terminal (a window where you type instructions to your computer). Open any terminal on your laptop, type claude, and the agent starts. The CLI is light: it adds Claude Code to a tool you already use, rather than being a tool of its own.
  • Claude Code Desktop. A standalone app. Open it like any other program. It comes with its own file browser, code editor, and terminal panel built in. Behind the scenes it runs the same engine as the CLI, so the two share the same settings.

The choice: do you want the agent inside an editor you already use, or an app that is only the agent? The CLI is the first answer. Desktop is the second. They aren’t exclusive, both can be installed with the same sign-in.

This journal runs on the CLI, inside the Cursor editor. The terminal inside Cursor runs claude, and that is the whole setup. Desktop sits on the same laptop as a side app for one specific feature called Dispatch (where a phone can send tasks for the laptop to run), not as the everyday tool.

Where the CLI lives

With the CLI installed, you need a terminal open somewhere to run claude in.

Installing Claude Code with CLI

If Cursor isn’t something you use, there is no reason to install it just for this. Any terminal works:

  • Your operating system’s built-in terminal. Terminal on macOS, Windows Terminal on Windows, the default terminal on Linux. Almost certainly enough to start with.
  • VS Code. A free editor from Microsoft with a built-in terminal, similar layout to Cursor.
  • Warp. A terminal with built-in helpers. Overkill until typing commands becomes a daily habit.

The CLI needs Node.js to run (a popular piece of software that lets programs run on your computer). First, check if Node.js is already on your computer. Open a terminal and type:

node --version

If a number appears (something like v22.11.0), Node is installed and you can skip to the next step.

If instead the terminal says command not found or similar, Node isn’t installed yet. Go to nodejs.org and download the LTS version (that stands for “long-term support”, the one recommended for most people). Run the installer like any other app. Once it finishes, close the terminal and reopen it (Node needs a fresh terminal to be recognised), then run node --version again to confirm.

With Node installed, install Claude Code with one command:

npm install -g @anthropic-ai/claude-code

Desktop path

Download the installer from claude.com/download. Run it like any other app. No Node, no extra setup.

Starting Claude Code

CLI. Open the project folder in your editor (Cursor, VS Code, or any other). Open the built-in terminal from the Terminal menu, or with the shortcut Ctrl+backtick (Windows and Linux) or Cmd+backtick (macOS). The terminal opens at the project’s root. Type claude and press Enter.

If you’re in a standalone terminal rather than one inside an editor, the sequence is the same with one extra step: cd into the project folder first, then type claude. For example on Windows:

cd C:\Users\yourname\projects\my-website
claude

On macOS or Linux the path shape is different, something like ~/projects/my-website.

Desktop. Launch the Claude Code Desktop app. On first run, pick “Open project” and point it at a project folder. The agent is ready from that moment.

First run on either path asks you to sign in. Both use the same Anthropic account, so you sign in once per device.

What Claude Code asks permission for

Every meaningful action Claude Code takes, it asks permission for first. Reading a file, saving a file, running a command on your computer, calling an online service. Each one appears as a small prompt: Allow this time? Allow always for this project? Deny?

This is on purpose. Claude Code can do anything you can do on the computer yourself, so a mistake can be as big as any mistake you could make by hand. Permissions are the safety net.

Most projects have a handful of actions that happen all day long: reading files you’ve already shared, saving small changes, running routine setup commands. Those are worth pre-approving, so the flow isn’t interrupted every thirty seconds. A smaller set of actions is worth checking manually every single time, because the consequence of getting them wrong is large.

A sensible starting point:

Pre-approve:

  • Reading files inside the current project folder.
  • Saving changes to files inside the project folder.
  • Running the routine setup commands your project needs (the ones you would be typing dozens of times per day). If you aren’t sure what those are yet, just add them when you see the prompt.

Never pre-approve:

  • Deleting files or folders.
  • Publishing code anywhere others can see it (pushing to GitHub, putting a site live).
  • Sending anything to a database real people use.

Even with a pre-approval list in place, Claude Code will pause when it tries to do something outside it. That pause is the safety net working.

The setup on screen

Two windows, full screen, side by side.

  1. Cursor, with the project folder open and the terminal inside it running claude.
  2. A web browser, for previewing whatever is being built: the live site, a local preview, a dashboard.

That is the whole setup. No plug-ins, no AI autocomplete, no elaborate configuration. Claude Code does the building; the browser shows the result.

The honest limits

  • It costs money. Short sessions are cheap. A full day of building can add up to a few euros or dollars. There is a monthly plan (called Max) for heavy use. Budget for it like any paid online service, because that’s what it is.
  • It can’t read your thoughts. The more you prepare before starting (the research you’ve done, the steps you’ve thought through, what you find aesthetically pleasing, what the goals actually are), the more Claude Code can deliver.
  • It can be confidently wrong. Check the changes it proposes before saving them into your project. “Claude said it tested this” is not the same as you testing it yourself.
  • It doesn’t magically know your project. The better organised the files are, the more useful Claude Code becomes. A later post in this series covers a simple way to keep Claude reading only the files it actually needs.

Try before you pay

If you want to try this setup before committing to a paid plan, I have a few Claude Code passes to share. Each gives a week of full access at no charge during the trial. DM me on Instagram. First come, first served.

What’s next in the series

  • One Hostinger VPS, every project. The rented server everything on this site runs on, with plain-English explanations of how a laptop reaches it, how several projects live on the same machine without interfering with each other, and how Claude Code fits into the whole picture.
  • Markdown pyramid. A small add-on (skill) for Claude Code that keeps it from reading every note in a project on every session.
  • …and more tools ahead, each tied to a real project as its receipt.

Subscribe below to get each one when it’s published.