Mwangaza
Getting Started

Getting Started with Claude Code

Kye · · 7 min read

Claude Code is an AI assistant that works inside your computer’s terminal. You describe what you want in plain English. It reads the files in your folder, writes new ones, runs commands and checks its own work. You never copy and paste code between a chat window and an editor.

By the end of this guide you will have Claude Code installed and a working budget tracker that opens in your browser. The tracker is small on purpose. The point is to see the whole loop once: describe, build, check, refine.

What Claude Code is, in one picture

Think of a capable assistant sitting at your desk with access to one folder on your computer. You tell them what you need. They open the files, make the changes, try the result and report back. They ask before doing anything that touches your system.

That is close to how it works. You start Claude Code inside a project folder, and that folder becomes its workspace. It can read everything in it, create and edit files, and run commands such as starting a program or installing a package. By default it asks your permission before editing files or running commands, so you see every action before it happens.

The difference from a chat tool like claude.ai is where the work lands. In a chat, you get code back as text and move it into place yourself. With Claude Code, the files appear in your folder, already saved.

What you need

  • A computer running macOS, Linux or Windows.
  • A paid Claude plan (Pro or Max) or an Anthropic Console account with API credit. The free plan does not include Claude Code.
  • A terminal you can open, and the ability to move between folders with cd. That is all the command-line knowledge this guide assumes.

You do not need to know how to program. It helps to be able to read a little code, and you will learn that faster by watching Claude Code work than from any course.

Step 1: Install Claude Code

On macOS or Linux, open a terminal and run:

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

On Windows, open PowerShell and run:

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

If you already have Node.js 18 or newer, the npm package also works:

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

Check that it installed:

claude --version

You should see a version number. If the terminal says command not found, close it, open a new one and try again. The installer updates your settings, and old terminal windows don’t see the change.

Step 2: Make a project folder and start Claude Code

mkdir budget-tracker
cd budget-tracker
claude

The first time you run claude, it opens a browser window to sign in with your Claude account. After that, you land at a prompt where you type instructions in plain English.

Always start Claude Code from inside the folder you want it to work in. It treats that folder as the project.

Step 3: Describe what you want

Paste this prompt:

Try this prompt: “Create a simple budget tracker web app. It should have a form to add expenses with a description, amount and category. Show a table of all expenses and a total at the bottom. Use HTML, CSS and plain JavaScript, with no external libraries. Keep it in a single index.html file so I can open it by double-clicking.”

Claude Code will explain its plan, then ask permission to create index.html. Read what it proposes and approve it.

Open the folder and double-click index.html. You should see a form and an empty table in your browser. Add two or three expenses and check that the total is right. Do the sum yourself, at least this once.

Step 4: Refine it

The first version is rarely the one you keep. Claude Code remembers the conversation, so ask for changes in the same session:

Try this prompt: “Add a breakdown by category under the table, showing the total for each category and its share of the whole as a percentage. Keep everything in index.html.”

Refresh the page in your browser. The breakdown should appear under the table.

Two more to try:

  • “Save the expenses in the browser so they are still there after I close the page.”
  • “Add a delete button to each row.”

Each change takes a minute or two. After four or five of these you will have a tool that does exactly what you want, which no off-the-shelf app quite does.

Three habits that make the difference

Give it context

Claude Code knows nothing about you or your work until you tell it. “Build a dashboard” gets you a generic dashboard. Compare:

Try this prompt: “I work for an NGO in Kampala. I have a CSV file called beneficiaries.csv with columns for month, district and number of beneficiaries. Build a single-page dashboard that shows the total, a breakdown by district, and a trend over the last 6 months. Use Chart.js for the charts.”

Who you are, what data you have, what you need to see and what tools to use. That one paragraph saves several rounds of corrections.

For context you want in every session, type /init inside Claude Code. It reads your project and writes a file called CLAUDE.md, which Claude Code reads each time it starts. Add your own notes to it: who the tool is for, the rules it must follow, what not to touch.

Refine, don’t restart

When a result is close but wrong, say exactly what is wrong. “The total counts the same expense twice when I edit a row” gets a fix. “It’s broken, try again” gets a guess.

Start a fresh session (type /clear) only when you switch to a different task. A long conversation about one thing is useful context. A long conversation about five things is noise.

Check the work

Claude Code is good, and it still makes mistakes. It can write code that runs perfectly and calculates the wrong number. You are the one who knows what the right number is.

Test with real data, including awkward cases: an empty form, a negative amount, a description with an apostrophe in it. Ask Claude Code to explain any part you don’t follow (“Explain what the calculateTotal function does, line by line”). You will learn more from those explanations than from a tutorial, because the code is about your problem.

About the permission prompts

By default, Claude Code asks before it edits a file or runs a command. On a small project that’s a handful of approvals. On a big build it becomes constant. When I built TARA, I was up until morning pressing “approve” on safe commands like python app.py instead of thinking about whether the numbers were right.

Resist the urge to switch the checks off entirely. Read what each prompt asks for your first few projects. You will quickly see which commands are routine. When you’re ready, the TARA case study shows how I used hooks and a settings file to approve the safe commands automatically and block the dangerous ones.

Three more things to build

Once the budget tracker works, try one of these in a new folder:

Summarise a spreadsheet.

Read sales_data.csv and write a one-page HTML report: total sales
by region, the top 5 products, and the monthly trend. Point out
anything unusual in the data.

Document something you already have.

Look at the files in this folder and write a README.md that
explains what this project does, how to set it up and how to
use it, with examples.

Automate a chore.

Build a small Node.js script that takes a folder of photos,
resizes them to 800px wide and saves the copies in a new folder
called "optimized". Leave the originals untouched.

Each of these fits in an evening and produces something you can use the next morning.

What to do this week

Pick one task from your own work that you do by hand every week. A report you assemble, a list you clean up, a calculation you redo in a spreadsheet. Open a new folder, start claude, and describe that task as if you were handing it to a new colleague.

The budget tracker teaches you the loop. The task from your own work is where Claude Code starts paying for itself.

Subscribe to Mwangaza to get the next guide when it’s published.

How this was written: Drafted with AI (Claude), edited and verified by the author.