Claude Code Setup Guide: How to Get Started Without Breaking Your Bank (Or Your Files)

• It's an agent, not a chatbot. Claude Code lives in your terminal and can edit files, run commands, and fix errors directly on your machine.
• Billing is separate. You need a paid Anthropic Console account (API credits) to run it, which is different from your $20/month Claude Pro subscription.
• Planning saves money. The /plan command is your most important tool for preventing "token burn" and keeping costs low.
• Systems beat syntax. You don't need to know complex code, but you do need to give clear, structured instructions in a CLAUDE.md file.
• Safety first. This guide covers how to set up spending limits and permission guardrails so you never accidentally delete important work.

Claude Code is an "agentic tool" developed by Anthropic. Unlike the web chat, which is trapped in your browser, Claude Code has hands. It can reach into your file system to read documents, edit content, run analysis scripts, and organize folders.

The core value proposition here is autonomy. In a web chat, you are the bottleneck — you have to move information back and forth. With Claude Code, you move from being the "doer" to the "manager," where you set the objective and the agent executes the loop until the task is done or it needs your approval.

The agentic loop works like this: you give a high-level goal (e.g., "Audit these 50 markdown files for broken links"), Claude Code proposes a step-by-step plan, executes it by reading files and making edits directly, then surfaces the results for your review. The heavy lifting happens between the plan and the review — freeing you from the copy-paste cycle entirely.

The "Cursor vs. Claude Code" Confusion

Professionals often ask if they need this tool if they already use Cursor. The answer depends on your goal. Community consensus is clear: Cursor is "autocomplete on steroids" for writing code, while Claude Code is an "autonomous employee" for completing tasks.

If you want to type faster, use Cursor. If you want to offload a repetitive 3-hour content migration, use Claude Code.

Before you open your terminal, let's make sure you have the right keys to the building. You cannot run Claude Code with just a standard Claude Pro subscription.

The Mandatory Checklist:

1. Anthropic Console Account: Sign up at console.anthropic.com and add credits (usually a $5 minimum). This is separate from your Claude.ai "Pro" account.
2. Node.js 18 or higher: This is the engine that runs the software.
3. Terminal Access: You need to be comfortable opening Terminal (Mac) or PowerShell (Windows).

Quick readiness check: Open your terminal and type node -v. If you see a version number like v18.x.x or v20.x.x, you're ready to install. If you see an error, go to nodejs.org and download the "LTS" version first.

Once you have your prerequisites, the installation is a single command. Experienced users verify that this process is generally safe, but you should always pay attention to the permissions prompts.

Step 1: Run the Install Command

Open your terminal and paste the following command. This tells the Node package manager (npm) to download the tool globally (-g) to your computer.

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

Step 2: Authenticate

Once the installation finishes, type:

claude

The terminal will ask you to press Enter to open your browser. This takes you to the Anthropic login page. Log in with the account that has your API credits.

Step 3: Handle the Permissions Loop

This is where new users often get confused. Claude Code is designed with a "Human-in-the-Loop" security model — it will ask for permission before it does anything significant, like reading a file or running a command.

You will see prompts like:

Allow Claude to run this command? (y/n)

Don't panic. This is a feature, not a bug. It ensures the agent doesn't accidentally delete your project. As you get more comfortable, you can streamline this, but for your first run, simply type y and hit Enter to approve actions.

When you first install Claude Code, it runs in "Guest" mode. It doesn't know your rules, your tone, or your preferences. This leads to the infamous "Context Compaction" problem where the agent forgets what you said 20 messages ago.

The solution is to give the agent a brain. This is called the CLAUDE.md file.

Step 1: Initialize Your Project

Go to the folder where you want to work. Type:

/init

This creates a CLAUDE.md file in your directory. Think of this file as the "Read Me First" document for the agent. Every time you start a new session, Claude Code reads this file to understand the project.

Step 2: Configure the Brain

Open CLAUDE.md in any text editor (VS Code, Notepad, TextEdit). This is where you define the "Rules of Engagement."

Here's a simple example of what a CLAUDE.md file looks like:

• Always verify broken links before deleting them.
• Use American English spelling.
• Do not modify the /archive folder.

• Be concise and direct.
• Avoid robotic phrases like "I hope this helps."

This file prevents "hallucinations" by giving the agent a single source of truth. If the agent makes a mistake, don't just correct it in chat — add a rule to `CLAUDE.md` so it never happens again.

### Step 3: Handle Global vs. Local Settings

Claude Code also uses a `settings.json` file to manage permissions. A common frustration reported by the community is that editing the global settings file often gets ignored.

Instead, rely on local project settings. When you run `/init`, a local configuration is created. This ensures your permissions apply specifically to this project, reducing the risk of "permission fatigue" across different workspaces.

You don't need to memorize a dictionary of commands. You likely only need four. Here is your cheat sheet for controlling the agent.

The Power of /plan

This is the most critical command. Experienced practitioners agree: never ask the agent to "just do it." Always ask it to /plan first.

When you type /plan Audit the blog for SEO, the agent will pause and list out the steps it intends to take. You can review this list, modify it, and approve it. This prevents the agent from getting stuck in a loop and burning through your API credits on a mistake.

Our Claude Code settings tutorial walks through configuring these commands and setting up your environment step-by-step, including how to customize which tools the agent is allowed to use.

You have the tool installed. You have your CLAUDE.md file set up. Now comes the mental shift.

Most users treat AI like a faster typewriter. They prompt, wait, review, and prompt again. But with Claude Code, you are no longer the typist — you are the manager.

This shift can be disorienting because you might feel the urge to "just do it yourself" since explaining the task feels slower than typing. But remember the goal: scalability. Once you teach the system how to handle a task, it can repeat that task 100 times without fatigue.

The real skill isn't knowing the command line flags. It's knowing how to structure instructions so the agent can execute them reliably. The Claude Code tutorials on this site are built around exactly this — each one teaches you a different pattern of structured thinking that translates directly to better agent output.

A great starting project: build your own custom Claude Code skill. You'll go from writing an instruction file to having a reusable command the agent recognizes by name — the same pattern professionals use to productize repeatable workflows.

Let's address the biggest concern: money.

There is a dangerous misconception that your $20/month Claude Pro subscription covers Claude Code. It doesn't.

Claude Code runs on the Anthropic API, which is a pay-as-you-go service. Every time the agent reads a file, thinks, or writes code, it costs money.

The Agent Teams Trap

A recurring warning from power users is to be extremely careful with "Agent Teams." This is an advanced feature where multiple AI agents work together on a single task. While powerful, it can burn through $50 of credits in minutes if a loop goes wrong.

Safety Recommendations:

1. Set a Spend Limit: In your Anthropic Console, set a hard monthly budget (e.g., $20).
2. Avoid Teams Initially: Stick to single-agent workflows until you are comfortable with the cost patterns.
3. Use /cost Frequently: Check your spend after every major task.

Getting Claude Code installed is just the first step. The real power comes from building a library of "Skills" and robust CLAUDE.md rules that turn the agent into a custom coworker.

You don't need to be a software engineer to do this. You just need to be clear, structured, and intentional. The same skills that make you a great UX writer or Product Manager — empathy, logic, and clarity — are the exact skills you need to master this tool.

Start small. Run a simple audit. Check your costs. And slowly, you'll find yourself doing less of the "doing" and more of the "directing." Browse the tutorials here to pick your first project and see what 20 minutes of structured prompting can produce.

Does Claude Code cost money?

Yes. Claude Code uses the Anthropic API, which charges per token (roughly per word). This is billed separately from your $20/month Claude Pro subscription. You must add credits to your Anthropic Console account to use it.

Can I use Claude Code on Windows?

Yes. You can use PowerShell or the Command Prompt. However, you must install Node.js for Windows first. Users often recommend using Windows Subsystem for Linux (WSL) for a smoother experience, but it's not strictly required.

How do I keep my session alive?

If you close your terminal, the active session ends. Advanced users often use a tool called tmux to keep sessions running in the background, but for beginners, simply keep your terminal window open until the task is complete.

What if I get a 529 error?

A 529 error usually means the Anthropic servers are overloaded. This is common during peak hours. Wait a few minutes and try running your command again. It's rarely an issue with your local setup.

Is Claude Code better than Cursor?

They serve different purposes. Cursor is an "AI-powered text editor" designed to help you write code faster. Claude Code is an "autonomous agent" designed to perform multi-step tasks for you. Most professionals use both — Cursor for writing, Claude Code for heavy lifting.