The Basics

Chapter 1 | The Basics Written for: newcomers who have never used Claude Code — covering every level

How to install Claude Code

Updated September 2026 The recommended way to install has moved from npm to the native installer. The difference that matters most: the native installer updates itself automatically in the background, while Homebrew and WinGet do not — you have to trigger those yourself.

What you need first

Requirement	Details
A terminal	Terminal on macOS/Linux, or PowerShell/CMD on Windows
A project	A code folder for Claude to work in
An account	Claude Pro, Max, Team or Enterprise / a Claude Console account / or through a supported cloud provider

Node.js is no longer required, and neither is an API key up front — not on the recommended route.

Route 1 — Native installer (recommended)

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash
 
# Windows PowerShell
irm https://claude.ai/install.ps1 | iex
 
# Windows CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

If you get The token '&&' is not a valid statement separator, you are in PowerShell, not CMD. If you get 'irm' is not recognized, you are in CMD, not PowerShell. The quick tell is the prompt: PowerShell starts a line with PS C:\, while CMD shows just C:\.

On native Windows, install Git for Windows as well so Claude Code can use the Bash tool. Without it, PowerShell becomes the shell instead. If you are on WSL, you do not need Git for Windows.

Route 2 — Package manager

# macOS — Homebrew
brew install --cask claude-code
 
# Windows — WinGet
winget install Anthropic.ClaudeCode

Homebrew has two casks. claude-code follows the stable channel, which lags by about a week and skips releases with serious problems. claude-code@latest gets each new release the moment it ships.

Neither of these updates itself. You have to run brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode yourself.

On Debian, Fedora, RHEL and Alpine you can install through apt, dnf or apk as well.

Check the install worked

claude --version

The command prints a version number followed by (Claude Code).

Step 2 — Sign in

Claude Code needs an account before it will run. The simplest route is to run claude on its own and let it take you to browser authentication.

claude

To switch accounts or sign in again later, type /login in an open session.

Accounts that work:

• Claude Pro, Max, Team or Enterprise (recommended)
• A pre-paid Claude Console account, which creates a workspace named “Claude Code” automatically on first login so all your spending is tracked in one place
• Enterprise cloud providers: Amazon Bedrock, Google Cloud Agent Platform and Microsoft Foundry
• A Claude apps gateway your organisation runs itself, where the administrator sets the URL for you and /login opens the Cloud gateway page for signing in with your organisation’s SSO

If ANTHROPIC_API_KEY is set, Claude Code skips the login page and asks you to approve that key instead. This route still works, but it is no longer the recommendation for ordinary use — signing in with an account is easier to manage and leaves no key sitting on your machine.

Once you have signed in successfully the credential is stored, so you never sign in again.

3	Start working Open a terminal, go to the project folder, and run claude

cd /path/to/your/project
claude

Steps: Windows (PowerShell)

Windows supports Claude Code through PowerShell directly, or through WSL2.

# Install (no need to run as Administrator)
irm https://claude.ai/install.ps1 | iex
 
# Sign in with an account — the recommended way
claude
 
# If you genuinely need an API key, set it for this session only
$env:ANTHROPIC_API_KEY = "sk-ant-api03-xxxxxxxxxxxx"
 
# Set it permanently in the PowerShell profile
Add-Content $PROFILE '$env:ANTHROPIC_API_KEY = "sk-ant-api03-xxxx"'

💡 Install Git for Windows too
It lets Claude Code use the Bash tool. Without it, PowerShell becomes the shell instead. Download it at git-scm.com/downloads/win

Route 2 — Installing through WSL2 on Windows

WSL2 (Windows Subsystem for Linux) gives the closest thing to Linux you can get on Windows. Recommended for developers working with a Linux stack

# 1. Open PowerShell as Administrator and install WSL2
wsl --install
 
# 2. Restart the machine, then open Ubuntu from the Start Menu
 
# 3. Update packages
sudo apt update && sudo apt upgrade -y
 
# 4. Install Claude Code (Node.js is no longer a prerequisite)
curl -fsSL https://claude.ai/install.sh | bash

💡 Reaching Windows files from WSL2
Windows files live under /mnt/c/Users/name/ — for example cd “/mnt/c/Users/John/Projects/myapp” and then run claude

Route 3 — Installing in a Docker container

Suited to teams that want one standard environment, or to use in a CI/CD pipeline.

# Example Dockerfile
FROM debian:bookworm-slim
RUN apt-get update && apt-get install -y curl ca-certificates git     && curl -fsSL https://claude.ai/install.sh | bash
WORKDIR /workspace
 
# Build and run
docker build -t claude-code .
docker run -it -e ANTHROPIC_API_KEY="sk-ant-..." \
  -v $(pwd):/workspace claude-code claude

An example install on VS Code

VS Code is the most popular code editor there is, and Claude Code works with it very well — through both the built-in terminal and an extension.

Method 1 — Through the integrated terminal (the simplest)

No extension to install. Use the terminal VS Code already has.

1	Open a terminal in VS Code Press Ctrl+` (backtick) or go to View → Terminal

2	Navigate to the project folder If you opened the folder through VS Code, the terminal is already there

# Check you are in the right folder
pwd
 
# Start Claude Code
claude

3	Start working together Give Claude instructions in the terminal below while you edit code in the window above

💡 The split terminal trick
Press Ctrl+Shift+5 to split the terminal into two panes — run Claude Code on the left, run your program or watch logs on the right

Method 2 — Install an extension

Extensions exist that make Claude Code smoother to use. Search the VS Code Marketplace:

Extension	Publisher	Notable feature
Claude Code (Official)	Anthropic	Connects to Claude Code directly, with a sidebar UI
Continue	Continue.dev	Supports several LLMs, Claude among them, through the API
Cursor	Cursor	A separate IDE with Claude built in

Installing the extension

Press Ctrl+Shift+X to open the Extensions panel

Type “Claude” in the search box

Pick the extension you want and press Install

Press Reload Window so the extension takes effect

Open the extension and enter your API key in its settings

💡 Use claude —add-dir
Run claude —add-dir /path/to/project to give Claude a folder outside the current working directory

Model selection

Claude Code supports several models, so you can pick the one that suits the job. Choosing the right model saves money and gets more done.

Supported models and what they are for

Use an alias rather than a full model name. An alias always points at the newest model in that line, so you write it once and it does not go stale when a new release lands.

Alias	Suited to
fable	The hardest and longest-running work
best	The newest model your account can reach
opus	Work that needs complex reasoning
sonnet	Everyday coding
haiku	Simple work repeated a lot
opusplan	Opus while planning, then a switch to Sonnet for the work itself
sonnet[1m] / opus[1m]	Opens a 1-million-token context window for long sessions and large codebases

If you really do want to pin a version, the full names supported as of September 2026 are claude-fable-5-1, claude-fable-5, claude-opus-5, claude-sonnet-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-sonnet-4-6, claude-sonnet-4-5 and claude-haiku-4-5

Pricing The official documentation does not publish per-token prices on the model settings page, but prices do appear in the /model picker when Claude Code talks to the Anthropic API directly. Outside providers such as Amazon Bedrock set their own prices, so those rows in the picker show no price at all — do not take a price from any guide; read it from the real picker or from the provider’s pricing page

How to choose a model

Method 1: Specify it on the command line

# Use the account default
claude
 
# Specify with an alias
claude --model opus
claude --model sonnet
claude --model haiku
 
# Open the 1-million-token context
claude --model sonnet[1m]

Method 2: Set an environment variable

# The recommended variable (needs Claude Code v2.1.236 or later)
export ANTHROPIC_DEFAULT_MODEL="sonnet"
 
# Add it to .zshrc or .bashrc to make it permanent
echo 'export ANTHROPIC_DEFAULT_MODEL="sonnet"' >> ~/.zshrc

The older ANTHROPIC_MODEL still works. The order of precedence is --model first, then ANTHROPIC_MODEL, then the model value in a settings file — including whatever /model saved — and finally the organisation default. ANTHROPIC_DEFAULT_MODEL takes effect only when none of the above sets anything.

Method 3: Change model mid-session

While Claude Code is running, type /model to see the models and switch straight away.

💡 A rule of thumb for choosing
Always start with Sonnet. Move to Opus only when Sonnet cannot answer, or the work is genuinely complex. Use Haiku for batch processing and CI/CD work ⚠️ Watch out for model looping
If Claude repeats itself or gets stuck on the same problem, switch to Opus for a while. Opus reasons more deeply and often solves what Sonnet is stuck on

Configuring your editor

Claude Code can connect to the code editor you already use, so it opens files and shows diffs (the differences in the code) automatically

Set the default editor

# See the current editor
claude config get editor
 
# Set it to VS Code
claude config set editor "code"
 
# Set it to Cursor
claude config set editor "cursor"
 
# Set it to Vim
claude config set editor "vim"
 
# Set it to Neovim
claude config set editor "nvim"

Configuring through settings.json

The main settings file is at ~/.claude/settings.json and you can edit it directly.

{
  "editor": "code",
  "model": "sonnet",
  "autoApproveEdits": false,
  "theme": "dark",
  "notifications": true
}

editor	The program that opens files, such as code, cursor, vim, nano
model	The default model, which can be given as an alias such as sonnet
autoApproveEdits	true = approve edits automatically (careful!), false = ask every time
theme	UI theme colour: dark or light
notifications	Show a notification when work finishes

Configuring the diff view

When Claude edits a file it shows you a diff of the change before you approve it, and you can have that diff open in your editor.

# Open diffs in VS Code
claude config set diffTool "code --diff"
 
# Or use vimdiff
claude config set diffTool "vimdiff"

Giving files to Claude Code

Claude Code can read and analyse files in several ways, from naming a path directly to handing over several files at once.

Method 1 — Reference a file with @ in the prompt

The simplest and most direct way. Type @ followed by the file path.

# Read one file
> explain the code in @src/utils/auth.js
 
# Read several files at once
> compare @old_api.py with @new_api.py and tell me how they differ
 
# Read and give an instruction at the same time
> fix the bug in @src/components/Login.tsx where validation does not run

Method 2 — Reference file paths in a one-shot prompt

Correction (updated 2026-09) The previous edition told you to use a --file flag. That flag does not exist. Put the path in the prompt text instead, and allow the Read tool.

# One file
claude -p "summarise report.pdf for me" --allowedTools "Read"

# Several files
claude -p "review the code in api.py and test_api.py" --allowedTools "Read"

Method 3 — Pipe content straight into Claude

# Send file contents through a pipe
cat error.log | claude -p "analyse these errors"
 
# Send the output of another command
git diff HEAD~1 | claude -p "write a commit message for me"
 
# Send output from a command
npm test 2>&1 | claude -p "fix the failing tests"

Supported file types

Type	Extensions	Notes
Code	.py .js .ts .go .java .cpp .rs .rb	Every programming language is supported
Documents	.md .txt .rst .csv	Can be read and analysed
Config	.json .yaml .toml .env	Can be inspected and edited
Logs	.log .out	Errors can be analysed
Images	.png .jpg .gif .webp	Claude 3+ can read them (vision)

💡 Hand over a whole folder
Type @src/ to have Claude read every file in the folder, but be careful with very large folders — they can fill the context. Name only the files that matter

Having Claude Code search the internet

Claude Code can search the internet through an MCP browser tool and web search, which keeps the information it works from current.

Method 1 — Web search through MCP

Install an MCP browser server first; after that Claude searches on its own whenever you ask.

Installing an MCP browser server

# The recommended way: let claude write the settings file for you
claude mcp add --transport stdio browser -- npx @playwright/mcp

# If you write it yourself, put it in .mcp.json at the project root
# (MCP is not configured in settings.json — see the scopes section in Chapter 2)
{
  "mcpServers": {
    "browser": {
      "command": "npx",
      "args": ["@playwright/mcp"]
    }
  }
}

Examples

# Once the MCP browser is installed, just ask
> find out what new features Next.js 15 has
 
> open https://docs.anthropic.com and summarise the latest API for me
 
> find a fix for the error "Cannot read properties of undefined" in React 18

Method 2 — Brave Search MCP

The Brave Search API gives good results and is more privacy-friendly. It needs a Brave Search API key.

# Add to ~/.claude/settings.json
{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": {
        "BRAVE_API_KEY": "BSA-xxxxxxxxxxxxxxxxxx"
      }
    }
  }
}
 
# Sign up for a Brave API key at brave.com/search/api/

Method 3 — Fetch a URL directly

If you want Claude to read one particular page, no extra MCP is needed — just ask.

# Ask Claude to fetch a URL directly
> read and summarise the content at https://react.dev/blog/2024/04/25/react-19
 
> compare the documentation at
  https://docs.python.org/3/library/asyncio.html
  with https://anyio.readthedocs.io/

💡 Check whether MCP is running
Run /mcp in Claude Code to see which MCP servers are connected. If browser shows green, it is ready to search

Having Claude Code search files on your machine

Claude Code can search and analyse files on your machine directly, through both built-in tools and an MCP filesystem server.

Searching with the built-in file tools

Claude Code ships with file-search tools; there is nothing extra to install.

# Find files by name
> find every file named config.json in this project
 
# Find code containing a specific word
> find the function named "getUserById" in this project
 
# Search by pattern
> find every file that imports from "@/components/ui"
 
# Analyse the structure
> show me the full folder structure of this project

Installing an MCP filesystem server (search the whole machine)

To search outside the folder you have open, or across several folders at once, install MCP Filesystem.

# Add to .mcp.json at the project root (not settings.json)
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/john/Projects",
        "/Users/john/Documents"
      ]
    }
  }
}
 
# List the folders you want Claude to be able to reach

MCP Filesystem examples

# Search across projects
> find every .env file in the Projects folder
 
# Compare projects
> compare the package.json of ProjectA and ProjectB
 
# Search for a pattern across the machine
> find the function named "sendEmail" in every .ts file under /Projects

Using ripgrep (rg) with Claude

ripgrep is a very fast text search tool, and you can use it with Claude through the Bash tool.

# Install ripgrep
# macOS: brew install ripgrep
# Ubuntu: sudo apt install ripgrep
 
# Ask Claude to search with rg
> use rg to find every "TODO" in this project and summarise them for me
 
> find every async function in src/ and review whether the error handling is complete

💡 Ask for something specific
Tell Claude exactly what you want found — “find every TODO that is still unfixed” or “find console.log calls someone forgot to remove” — and Claude picks the right tool itself

Driving Claude Code by voice

Claude Code accepts spoken instructions through speech-to-text, which makes the work easier — particularly when the explanation you need to type is a long one

Method 1 — Use your OS voice system

macOS: Dictation

1	Turn on Dictation Go to System Settings → Keyboard → Dictation and switch on Enable Dictation

2	Set a shortcut Choose a shortcut, such as pressing fn twice or holding fn down

3	Use it Click the Claude Code input field, press the shortcut, speak your instruction, and press Enter when done

# An example of something spoken and sent to Claude
> "write a function to validate an email address,
   make it support Thai domains too, and include unit tests"

Windows: Windows Speech Recognition

# Turn on Windows Speech Recognition
1. Press Win + H to open Voice Typing
2. Or search for "Speech Recognition" in Settings
3. Click the Claude Code window, press Win + H, and speak

Method 2 — WhisperKit or Whisper (recommended)

OpenAI Whisper is a very accurate speech-to-text model with good Thai support, and its output pipes straight into Claude.

# Install Whisper
pip install openai-whisper
 
# Install ffmpeg (required)
# macOS: brew install ffmpeg
# Ubuntu: sudo apt install ffmpeg
 
# Record audio and turn it into a prompt
# 1. Record audio (5 seconds)
rec -r 16000 -c 1 command.wav trim 0 5
 
# 2. Transcribe it and send it to Claude
whisper command.wav --language Thai --model medium | \
  grep -oP "(?<=text: ).*" | claude -p -

Method 3 — Use a voice input app

Some apps are built for voice-to-Claude specifically, and are easier than writing a script.

App / tool	Platform	How you use it
SuperWhisper	macOS	Press a shortcut, speak, paste the result
Whisper Dictation	macOS/Windows	A menu-bar app; press the hotkey and speak
Voice Control	macOS	Built into macOS, with Thai support
Talon Voice	Every platform	Advanced voice control for developers

An example voice + Claude Code workflow

Press the hotkey to open the microphone

Say: “build a login API endpoint using a JWT token, and write tests too”

The system transcribes the speech automatically

The text lands in the Claude Code input field

Press Enter to send it

💡 Speaking so it works
Speak slowly and clearly, pronounce technical loanwords in English — say “function” rather than a transliteration — and always give the context first, as in “in a Next.js TypeScript project, create…”

Glossary for this chapter

Node.js	A runtime for running JavaScript on your machine, outside a web browser
npm	Node Package Manager, the package manager for JavaScript
nvm	Node Version Manager, a tool for handling several Node.js versions
API Key	A secret code for reaching an AI service; keep it secret and never put it in code
WSL2	Windows Subsystem for Linux 2, which runs Linux on Windows
MCP	Model Context Protocol, the standard for connecting AI to external tools
Diff	The difference between two versions of a file, showing what was added, removed or changed
Context Window	The largest amount of information the AI can take in at once (200K tokens)
Speech-to-Text	Technology that turns spoken words into text
Whisper	A speech-to-text model from OpenAI, supporting many languages including Thai
ripgrep (rg)	A very fast text search tool, far faster than grep
Environment Variable	A system variable holding config values, read by a program at startup; safer than putting them in code
One-shot Mode	Running Claude Code once and done with claude -p, without opening an interactive session