How to install Claude Code and finish your first project
Claude Code installation, a guided first project, daily commands, working with git, and plans as of July 2026.
Claude Code is Anthropic's tool that lives in your terminal and works directly on your project: it reads your files, proposes changes, runs tests, and shows you every edit before applying it. This guide takes you from zero to a working first project, step by step, with the exact commands and the decisions worth making at each point.
By the end you will have Claude Code installed, an active session linked to your account, and a small project built with its help, plus the basic habits that make the tool pay off from day one.
What you need before starting
The list is short and you probably cover it already:
- A computer running macOS, Linux, or Windows. On Windows it works natively or through WSL.
- Any terminal: the macOS one, your favorite Linux emulator, or PowerShell and Windows Terminal.
- A Claude account. The Pro plan includes Claude Code access inside the subscription, and the Max plans are built for heavy use. A pay-as-you-go route through the API console exists as well.
- Node.js 18 or newer if you pick the npm install. The native installer skips that dependency entirely.
With that ready, installation takes a couple of minutes.
Installation
One common install route uses npm:
npm install -g @anthropic-ai/claude-codeAnthropic also ships a native installer for macOS and Linux that downloads the binary directly, handy on machines where you prefer to skip the Node dependency:
curl -fsSL https://claude.ai/install.sh | bashWhen it finishes, confirm everything landed in place:
claude --versionIf the command answers with a version number, the install is complete. The claude doctor command also checks your environment and flags any pending configuration detail.
First launch and linking your account
Move into any project folder, or create an empty one for this practice run:
mkdir my-first-project
cd my-first-project
claudeThe first time, Claude Code asks you to sign in. It takes you to the browser to authorize with your Claude account, and once you return, the terminal stays linked. From then on, every session inside that folder remembers the project context.
You will see a conversation interface inside the terminal: you type what you want in plain language and Claude Code answers, reads files, and proposes actions. Every delicate action, like modifying a file or running a command, asks for your approval before executing.
Your first project, step by step
For the first exercise, something small and visible works best: a to-do list web app in a single HTML file. It is enough to see the full work cycle with Claude Code.
### Step 1: ask for the first version
Type into the session:
Create an index.html file with a to-do list: text field, add button, and each task with a way to mark it as done and to delete it. Clean design, zero external libraries.
Claude Code generates the file and shows you a summary of its contents. Open the file in your browser and try it. You already have a working app.
### Step 2: iterate with concrete instructions
The key to working with Claude Code is asking for specific changes, one at a time:
Tasks marked as done should appear crossed out and drop to the end of the list.
Save the tasks to localStorage so they survive a page reload.
Add a counter on top: how many tasks exist and how many are done.
After each change, the tool shows exactly which lines it modified. Reload the browser and verify. This rhythm of asking, reviewing, and testing is the natural way to work: small changes you can review at a glance.
### Step 3: let it explore and explain
Claude Code also reads what already exists. Try questions like:
Explain how the function that saves to localStorage works.
What happens if I open the file in two tabs at once?
Answering this kind of question about real code is one of its strengths, and it serves you just as well on inherited projects or code written by someone else.
### Step 4: create the CLAUDE.md file
Run the command:
/initClaude Code analyzes the folder and generates a CLAUDE.md file with a description of the project, how to run it, and any conventions it detects. That file loads automatically in every future session, so the tool starts with context instead of asking you the basics each time. In larger projects, CLAUDE.md is the place to record how tests run, which style the code follows, and which folders are delicate.
Commands you will use daily
/helpshows the full command list./clearwipes the current conversation to start a fresh topic with clean context./compactsummarizes the conversation when it grows long, keeping what matters./costshows session consumption when you work through the API.claude -cresumes your previous session from the terminal.
Those five cover everyday use. You will discover the rest as you need it.
Work with git from the first change
Before asking Claude Code for anything on a project you care about, create the repository:
git init
git add -A
git commit -m "starting point"With a clean commit as your base, every change the tool proposes compares against something firm: git diff shows you the exact state, and rolling back costs a single command. Claude Code also understands history: you can ask it to review what changed since the last commit, draft the next commit message, or explain a specific line with its context. The combination of terminal, git, and Claude Code is where the tool feels complete.
How to ask for better results
The difference between a productive session and a frustrating one almost always lives in the instruction. Three habits that help:
One change at a time. "Add validation to the form" beats a list of ten changes at once. You review quickly and each step stays clear.
Concrete context. Mention the file, the function, or the console message that brought you here. "The save button in index.html stops responding after a reload" gives far more material than "something is off."
Review before accepting. Claude Code shows the proposed changes. Read them, especially early on: it is the best way to learn what the tool does and to keep the project under your control.
A fourth habit grows with practice: keep a running note of the prompts that worked well for you. Good instructions are reusable assets, and collecting them turns each session into training material for the next one.
Plans and limits, verified as of July 2026
At the time of writing, the official Claude pricing page lists the Pro plan at 20 dollars a month, with Claude Code included and usage limits shared with chat. The Max plans, at 100 and 200 dollars a month, extend those limits for heavy work. Since June 2026, interactive terminal use and programmatic use through the Agent SDK count against the same subscription limits, so checking the current pricing page makes sense as your consumption grows. The alternate route is the API console with pay-as-you-go billing, useful when you prefer variable spend.
For getting started, the Pro plan comfortably covers this guide and personal projects of moderate size.
First-day questions
Does it work in any folder? Yes. Open it at the root of any project and it reads what it finds: Python, JavaScript, Go, documentation, anything. The clearer the project structure, the better it works.
Do my files get uploaded somewhere? The tool sends the relevant content to the model so it can reason about it, as a normal part of how it operates. On projects with delicate data, review Anthropic's privacy documentation and tune the permissions to your needs.
Do I need to know how to program to use it? It helps, a lot, because reviewing the proposed code is part of the cycle. That said, plenty of people use it for terminal tasks, documentation, and file organization while they learn.
Can I use it for things beyond programming? It also organizes notes, drafts documentation, analyzes data, and automates terminal tasks. Programming is its strongest suit, and the rest comes from the same ability to read, reason, and act on files.
What comes after the first project
With the install and the first cycle under your belt, the natural path is bringing Claude Code to a real project of yours: open it in that folder, run /init, and start with a small, well-defined task, like adding a test or fixing a specific detail. The learning curve is climbed by working on code that matters to you. Team use is worth exploring too: sharing the project's CLAUDE.md with your team aligns everyone on the same setup, and running sessions on separate branches keeps the history tidy. One more habit worth building early is keeping sessions short and themed: a fresh conversation per task keeps the context sharp, and clearing between tasks costs a second.
The next topics worth exploring are the most-used commands and shortcuts, recommended practice projects for leveling up, and the extensions that connect Claude Code with more tools. Each one gets its own guide here.
Sources
Was this article helpful?