Notes: Installation and Setup

In This Section, You Will Master:

• Getting Claude Code installed and authenticated on your own machine, whichever platform you are on
• The prerequisites that decide whether an install goes smoothly or fails with an unclear error
• Choosing a plan with real numbers in front of you, and knowing what a token actually is
• Running your first session and reading what it cost
• Handing an error back to the tool that produced it, which is the habit you will reach for most

How to Use These Notes:

These notes are designed as quick-reference companions to your video learning.

• Use them to review key concepts after watching lectures
• Reference them when you need a quick reminder of frameworks and tools
• 🎯 Core Concepts = Essential terms and frameworks
• 🚀 Actionable Items = Practical implementation steps

The Growth Payoff:

You started with a blank terminal. By the end of this section Claude Code is installed and authenticated, a plan is picked with your eyes open, a real project you described in plain English exists on your machine, and you have a cost reading to compare everything else against. The debugging reflex you build here compounds with every problem you meet from now on.

---

Lecture: The Screen That Scares People (Before You Install)

Why beginner tutorials lose people in the first two minutes, and what is actually different between you and the person on screen.

🎯 Core Concepts:

• Tutorials lose people by moving between a terminal, an editor and a browser without ever saying which is which, or when they moved
• People stop watching before anything has been taught, and the reason is not that it went too fast
• Every other program tells you what it can do by showing you things to click. A terminal gives you a blinking line and nothing else
• That is not a test and nothing is being kept from you. It was built for people who already knew what to type, and never got a second design

---

Lecture: What will you accomplish in the next 60 minutes? (Your Setup Plan)

What is true on your machine by the end of this section, and how the install lectures are laid out.

🎯 Core Concepts:

• Three things will be true: Claude Code installed and authenticated, a project you described in plain English sitting in your folder, and a cost reading telling you what it took
• Every install path gets its own walkthrough, because the route genuinely differs across macOS, Windows and Linux
• The five most common errors have their fixes ready before you hit them

---

Lecture: What do I need before installing Claude Code? (Prerequisites)

The thirty second check that tells you how ready your machine already is.

🎯 Core Concepts:

• Two commands answer it: node --version and npm --version. Node is the runtime that lets your computer run programs written in JavaScript, npm is the package manager that installs them
• Your operating system should be reasonably current, with at least 4 GB of RAM and an internet connection
• On Windows, Git for Windows has to be installed first, or the install fails with an error that does not explain the cause
• A paid Claude plan is required. The free Claude.ai plan does not include Claude Code
• PATH is the list of folders your shell searches when you type a command. When something says add this to your PATH, it means teaching your shell where to look

🚀 Actionable Items:

• Run both version commands and write the numbers down, so you can compare against the finished install

---

Lecture: Which terminal should I use? (Warp vs the One You Have)

Your own terminal runs Claude Code perfectly well, and what a purpose-built one adds.

🎯 Core Concepts:

• Warp runs exactly what your own terminal runs, so nothing you learn is Warp only. What it adds is that most of it can be done by clicking
• It is free, and free here is not a trial that runs out
• The panel on the left moves you between folders by clicking, and the command history shows you the commands it wrote for you, so you have been using them all along

🚀 Actionable Items:

• Skip the sign in and skip Warp's own AI. The AI you came here for is Claude Code, and it runs inside any terminal
• Set Warp as your default terminal, so right-clicking a folder offers Open in Terminal
• Turn on Claude Code notifications. A real job can run four to ten minutes, and you are not meant to sit and watch it

---

Lecture: What is that folder actually doing? (Your Project Folder)

The folder you start in is a boundary, and choosing it on purpose is what lets you say yes quickly later.

🎯 Core Concepts:

• Whichever folder your terminal is sitting in when you start Claude Code is the only place Claude Code can write. That folder, and anything inside it
• So the folder is a fence, and you are the one who put it up. It was never about being tidy
• The line just before the cursor tells you where the fence is. One look before you start is the cheapest habit in this course
• Written in full a folder path starts at the top of your machine. The tilde is shorthand for your home folder. Same place, fewer characters
• A room is exactly the right amount to hand over. On the day something unexpected happens, the damage has a boundary and you know where it is

🚨 Red Flags:

• Do not put projects inside a folder that syncs to the cloud. Git rewrites thousands of tiny hidden files constantly, a sync app copies them halfway through being written, and then starts making duplicates. Nothing announces it. It looks like a project that worked fine for a week and then began behaving strangely

---

Lecture: How do I install Claude Code on macOS? (Install)

One line, then two checks that tell you it worked properly.

🎯 Core Concepts:

• The native installer is recommended because Claude Code refreshes itself whenever a newer version is available
• Homebrew is an alternative, and it does not auto-update, so you upgrade by hand
• Installing through npm is deprecated and no longer recommended
• claude --version confirms it worked. claude doctor runs a full diagnostic across your shell configuration and authentication
• The first time you run claude, your browser opens and you sign in once. The credentials cache for every future session

🚀 Actionable Items:

• On a company managed Mac, three routes usually get through: Homebrew, the Visual Studio Code extension, or the web version at claude.ai/code which needs no local install at all

---

Lecture: How do I install Claude Code on Windows with WSL2? (Install)

A full Linux environment inside Windows, so every command and config file you find online works as written.

🎯 Core Concepts:

• WSL2 gives you a real Linux system running inside a window on your Windows desktop, which is why tutorials that assume a Linux terminal just work
• wsl --install from PowerShell as Administrator does the whole setup, once
• Ubuntu's first launch creates a Linux user account separate from your Windows login. The password prompt shows nothing at all while you type, and that is normal
• From there, the install is identical to the Linux path
• If claude --version says command not found, the install worked and the terminal has not been told where to look yet. One line fixes it

🚀 Actionable Items:

• Install Windows Terminal while you are here, so you can tab between PowerShell and Ubuntu in one window

---

Lecture: How do I install Claude Code on Native Windows? (Install)

The simpler path, and the honest comparison so you can choose with confidence.

🎯 Core Concepts:

• Native handles most of the work in this course and in daily professional use, especially web development, API building and scripting
• WSL2 is the stronger choice if your work involves Docker, containers, or open source projects that assume Linux
• Git for Windows has to be in place first, because Claude Code uses Git Bash internally for certain file operations
• irm ... | iex is the PowerShell equivalent of the curl | bash pattern on macOS
• Native and WSL2 coexist on one machine without conflict, but they are separate installs and each needs its own authentication

---

Lecture: How do I install Claude Code on Linux? (Curated Walkthrough)

The same installer as macOS, and the three places Linux installs tend to go sideways.

🎯 Core Concepts:

• The same curl installer works on Ubuntu, Debian, Fedora, Arch and most other distributions, because it is universal across the family
• Watch three things: your Node version, PATH configuration, and the authentication step at the end
• The installer drops Claude into ~/.local/bin, which is already on PATH on most distributions but not on minimal setups
• The fix is one export line in your shell config file, then reloading it with source
• Running Claude Code inside Docker, a cloud VM or a remote server uses the identical process, so installing once teaches you every Linux environment

---

Lecture: How do I troubleshoot Claude Code installation errors? (Common Fixes)

The five errors that cover roughly ninety percent of what goes wrong, each with its fix.

🎯 Core Concepts:

• Git Bash missing on Windows, permission denied on macOS or Linux, bash not found on minimal container images, missing system libraries on Alpine Linux, and the npm versus native conflict
• Almost every one takes under sixty seconds to fix once you have seen it
• The native installer does not need sudo at all, so permission errors are not something to escalate
• The pattern across all of them: read the error carefully, and search for the exact message text

🚀 Actionable Items:

• For anything not on the list, paste the complete error message into an AI tool. The exact error text is the thing that matters

---

Lecture: What is an API, and what is a context window? (Before You Pay)

Two words that turn up inside the plan decision, both offered as reasons to spend more, and neither usually explained.

🎯 Core Concepts:

• An API is a channel one program uses to talk straight to another, with nobody clicking anything in between
• Claude has a channel of its own, and going in through it is a second way of paying for the very same Claude. Not a different tool, a different way of paying
• Two ways to buy: a plan with a set monthly amount and a limit you wait out, or the API channel with a published rate per amount of text and no ceiling at all
• The API sounds like the flexible one. For the work ahead of you it is the expensive one, so start on a plan
• The context window is the space holding everything in one conversation. When it fills up, the oldest part starts falling out of the back, which is why a long session forgets what you said at the beginning
• Nothing switches that channel on behind your back. Signing in with a plan uses the plan

---

Lecture: Which Claude subscription plan is right for you? (Pro vs Max vs Team)

The clean breakdown the pricing page does not give you, with honest notes on where each tier runs out.

🎯 Core Concepts:

• Almost everyone starting here should be on Pro. The question is when a higher tier starts making sense for your situation
• Max at one hundred gives roughly five times the usage for five times the price. Max at two hundred gives roughly twenty times for ten times, which is why daily professional users tend to land there
• When you hit a limit you have three options, not one: wait for the window to reset, upgrade, or enable API billing so the overflow keeps running at standard rates
• A token is a small chunk of text, often shorter than a full word. Every prompt you type, every file Claude reads and every line it writes costs tokens
• The single most expensive habit is mixing unrelated tasks in one session. The community calls it a kitchen sink session, and performance degrades as the context fills with things that are not relevant

🚨 Red Flags:

• If you enable API overflow, make sure any script handles rate-limit errors explicitly rather than blindly retrying. A single misconfigured loop has been reported to drain hundreds of dollars in minutes

🚀 Actionable Items:

• Open a fresh session whenever the task changes. That one habit saves more money than choosing the right plan ever will

---

Lecture: How do I run my first Claude Code session? (Hello World)

The smallest version of it works, built on your own machine, with a cost reading at the end.

🎯 Core Concepts:

• Type claude and press enter. The difference from the earlier demo is that this time you are driving
• Claude typically asks one or two clarifying questions first, often as multiple choice with one option marked Recommended. You can pick that, choose another, or type your own answer
• The same agentic loop from the first section runs, except your project is the one being built
• Three options on whatever comes back: accept it, reject it and ask for a different direction, or modify specific parts inside the same session
• /usage shows total cost, API time, wall time and the code changes for the session

🚀 Actionable Items:

• On a Pro or Max subscription, the dollar figure is not what you are billed. It is the API-rate equivalent of the work, useful as a value reference

---

Lecture: How do I build my own Hello World? (Exercise)

The same loop again, in your own words, with no demo to copy.

🎯 Core Concepts:

• Three options: a personal landing page, a to-do list app, or a calculator. If you cannot pick, take the landing page, because the muscle memory carries over directly
• Write your own prompt rather than copying a template. Part of what you are learning is how to talk to Claude Code, and your natural language works better for that than you might expect
• Your first attempt does not have to be elegant. Claude Code is forgiving and asks clarifying questions when a prompt is unclear
• What matters is running the full cycle yourself, from writing the prompt to reviewing the result

🚀 Actionable Items:

• Open a new terminal, type claude, give it your prompt, then review and decide. Run /usage when you are done

---

Lecture: How do I let Claude Code troubleshoot itself? (AI-Powered Troubleshooting)

The habit that turns a twenty minute browser hunt into a sixty second resolution.

🎯 Core Concepts:

• Hand the error back to the same tool that produced it
• Most of the time Claude already does this without being asked. When a command it ran returns an error, it reads it, names the cause, and offers a fix
• When the terminal has a lot of other output mixed in, copy only the specific error lines and paste them inside square brackets with your question
• It works because Claude already has the context. It knows what it was trying to do, what files are around, and what went wrong. A search result can never carry that
• The second-order effect is that you start experimenting more freely, because the cost of trying something unfamiliar drops to almost nothing

🚀 Actionable Items:

• The sequence is always the same: ask for a diagnosis, check the fix makes sense for your situation, then apply it

---

Lecture: That is day 2 done (Installed, Paid For, Running)

The question that decides whether today taught you anything, and the honest answer to it.

🎯 Core Concepts:

• A working page appeared on your screen and you did not write a line of it. So did you learn anything, or did you watch a machine work?
• Answering that wrong is where people quietly stop. It works, and it keeps working, right up until something breaks that they cannot fix
• Typing the code yourself is not what saves you from that. People assume it is
• What saves you is asking why it works, then changing one thing and watching what that does
• You already did all of it today. You chose a folder on purpose, wrote a prompt in your own words, and judged what came back

🚀 Actionable Items:

• Tomorrow is the first day you might bump into a plan limit. Nothing is lost and nothing is charged, so do not work carefully just to protect a number

This resource is part of

Claude Code: From Beginner to Pro Agentic AI Engineer

by Arun Nagarathanam

Take your skills to the next level with

AI GEO 2026: Get Recommended by ChatGPT, Google, Perplexity

by Arun Nagarathanam