Claude Code in the terminal is one install command, one claude, and then plain sentences. The part that stops people is not the agent; it is the ten minutes around it, where a shell says the command does not exist and the guide you found lists flags that were never in the official pages.
This page stays inside the official documentation: every command below is copied from the Claude Code terminal guide, quickstart, overview or CLI reference.
What Claude Code is, in the vendor’s own words
The overview opens with a description worth reading closely: “Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser.”
The terminal is one surface among several, and they share one engine: “Each surface connects to the same underlying Claude Code engine, so your repo’s CLAUDE.md files, settings, and MCP servers work across all of them.” If you already use the IDE extension or the desktop app, the terminal does not start from zero.
What the install expects from your machine
Three constraints are published, and checking them first saves a failed install. macOS needs version 13.0 or later. Windows needs version 10 build 1809 or later, and Claude Code does not support 32-bit Windows. Linux support depends on the distribution, which the terminal guide links to rather than lists.
On Windows, Git for Windows is optional but useful: it provides Git Bash, which is what the Bash tool uses. Without it, Claude Code runs through PowerShell.
One account requirement: a Claude subscription (Pro, Max, Team or Enterprise), a Claude Console account, or access through a supported cloud provider. If ANTHROPIC_API_KEY is already exported, the docs say Claude Code skips the login prompt and asks you to approve the key instead.
Install: one line per platform
# macOS, Linux and WSL
curl -fsSL https://claude.ai/install.sh | bash
# confirm the install
claude --versionThe Windows equivalents are a PowerShell line, irm https://claude.ai/install.ps1 | iex, and a CMD line, curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd. PowerShell and CMD are different programs with different commands, and the guide gives the tell: PowerShell lines start with PS C:\Users\YourName>, CMD lines do not.
From a package manager instead
Two package managers are documented. Homebrew installs the macOS build with brew install --cask claude-code, and WinGet installs the Windows build with winget install Anthropic.ClaudeCode. Linux package managers are mentioned for Debian, Fedora, RHEL and Alpine, with the commands living on the Linux install page rather than the terminal guide.
claude --version prints something like 2.1.211 (Claude Code), the only signal the docs give that the binary on your PATH is the one the installer wrote.
Start a session, and what the screen asks of you
cd /path/to/your/project
claudeRun it inside the repository you want to work on. There is nothing to upload: Claude Code reads your project files as needed. After launch the prompt shows the version, the current model and the working directory above it.
Logging in happens once, in a browser window the terminal opens for you. Inside a running session, /login does the same thing. Once you are in, the session commands published in the quickstart are three: /help shows what is available, /clear drops the conversation history, and /exit leaves, as does pressing Ctrl+D twice on an empty prompt. /resume reopens an earlier conversation.
The keys that replace the mouse
The terminal guide’s most practical paragraph is about not being able to click. Arrow keys move around, Enter sends a message and approves a permission request, and Esc interrupts Claude while it is working. Those four habits cover the first hour.
Shift+Tab is the one worth memorising. It cycles the permission mode of the session you are in, and the docs note that the starting mode depends on your plan.
Run one task without opening a session
A session is not always what you want. The CLI reference documents shorter forms that print an answer and return you to the shell.
# start interactive mode with an opening prompt
claude "fix the build error"
# answer one question, print it, exit
claude -p "explain this function"
# continue the most recent conversation in this directory
claude -c
# resume a specific session by ID or name
claude -r "auth-refactor" "Finish this PR"The distinction between claude "query" and claude -p "query" is in the reference and is easy to miss: the first opens interactive mode with a prompt already filled in, while -p answers and exits, which is what you want in a script. -c and -r both restore earlier work; -c takes the latest conversation in the current directory, -r takes the one you name.
Who approves a file edit
Nothing is written without a decision from you, and the published starting mode depends on the plan. The quickstart states that auto mode is the built-in starting permission mode for interactive terminal sessions on Pro, Max and Team plans, where a classifier reviews actions instead of you and Claude edits most files and runs most commands without asking. On other plans, Manual mode is what a session starts in.
Either way, Shift+Tab switches modes mid-session, and your settings or your organisation can set a different starting mode. When a prompt does arrive, Enter chooses Yes. The terminal guide puts it plainly: before creating or changing files, Claude asks for permission.
Anthropic’s pricing FAQ answers the billing question: Claude Code is included in all paid plans and shares the same usage limits, drawing from one pool. The plan breakdown walks the tiers.
When the command is not found
command not found: claude is the most common first-run failure, and it has one cause: the install directory is not on your PATH. The installer prints the exact fix at the end, and the guide repeats it for both shells.
# Zsh, the macOS default
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc
# Bash, the Linux default
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrcIf the installer returned a web page
HTML instead of a script, or syntax error near unexpected token, means the install URL answered with a page rather than the installer. The guide names the specific page to look for: “App unavailable in region” means the country or region is not supported.
Three more messages are documented. On macOS, a dyld error mentioning built for Mac OS X 13.0 means your OS is older than the supported version. On Windows, irm is not recognized means you are in CMD rather than PowerShell, and 'claude' is not recognized is the Windows spelling of the PATH problem, fixed by adding %USERPROFILE%\.local\bin. A message about 32-bit Windows means a PowerShell window launched from the (x86) entry, which you should close and reopen without it.
Frequently asked questions
Do I need to know how to code? The terminal guide says no, and suggests describing what you want in plain English: make me a simple webpage that says hello world.
Is the terminal required at all? No. The overview lists a terminal, VS Code and JetBrains extensions, a desktop app and the web at claude.ai/code. Terminal is the surface with the fullest command set.
How do I update it? Homebrew users run brew upgrade claude-code and WinGet users run winget upgrade Anthropic.ClaudeCode. The CLI reference also documents claude update, which updates to the latest version.
Where do I start once it runs? The Claude getting-started guide covers the product side, and if you would rather call the API directly than drive an agent, the API getting-started guide shows the request shape. A different kind of first step, comparing what you would pay for, is in the Claude vs DeepSeek comparison.
Install it, confirm the version, start it inside a repository, and press Shift+Tab once to see which mode you are in. Everything after that is sentences.