Claude Code isn’t a chat window in your browser. It opens in your terminal, reads your project’s files on its own, edits them, runs the tests and commits. Say “fix that bug” and the expectation is that it finds the file and changes it.
This guide walks the path from install to your first code change, and gives a straight answer on the free-version question 🧑💻
The Short Answer
- Install (Windows PowerShell):
irm https://claude.ai/install.ps1 | iex - Install (macOS, Linux, WSL):
curl -fsSL https://claude.ai/install.sh | bash - Start it:
cdinto your project and typeclaude - No free tier: you need a Claude subscription (Pro, Max, Team, Enterprise) or a prepaid Console account
- First move: run
/initto generate aCLAUDE.mdfor the project
What Is Claude Code?
Claude Code is Anthropic’s coding agent for the terminal. What separates it from a regular AI assistant is that it doesn’t only produce text. It has tools for reading files, writing files, running commands and searching the web, and it decides on its own which to use and in what order.
In practice the loop looks like this: you say what you want, Claude locates the relevant files, reads them, makes the change, runs the tests, looks at the output and fixes what’s broken. That loop is what “agentic” means.
The terminal isn’t the only surface. The same product ships as a desktop app, a VS Code and JetBrains extension, a browser version, and a GitHub Actions integration. This guide covers the terminal, because everything else runs on the same model.
Ask Claude on the web for code and it gives you the code. Copying and pasting it is your job.
Claude Code applies it. It finds the file, edits it, runs it and shows you the result. You don’t paste context either, it reads the project as it needs to.
Is Claude Code Free?
No. There is no free tier. To run it you need one of:
| Account type | How it works |
|---|---|
| Claude subscription (Pro, Max, Team, Enterprise) | The recommended route. Usage is included in the subscription, no per-token bill on top. Your allowance resets on a rolling five-hour window and a weekly one |
| Claude Console | Runs on prepaid API credit and bills per token. On first login a workspace called "Claude Code" is created for you automatically |
| Cloud providers | Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, for enterprise setups. The bill lands in your cloud account |
For current subscription prices, claude.com/pricing is the source that stays correct.
If you’re on the API side, Anthropic’s own docs give this enterprise average: roughly $13 per developer per active day, and $150-250 per developer per month. For 90% of users it stays under $30 per active day. On a subscription none of this applies, you pay a flat fee.
Search results turn up account-sharing offers and token resale listings promising free Claude Code. Those end in account suspension, and most of them are outright scams.
If budget is tight, the sensible route is a Console account with a small credit balance, billed per token. A few dollars is enough to try it properly.
Installation
One line. Pick the one that matches your OS.
Windows
Open PowerShell and run:
irm https://claude.ai/install.ps1 | iex
On CMD the command is different:
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd
Not sure which one you’re in? Look at the prompt. PS C:\ means PowerShell, plain C:\ means CMD.
If you prefer WinGet:
winget install Anthropic.ClaudeCode
Claude Code runs commands through its Bash tool. Without Git for Windows it falls back to PowerShell, and some commands won’t behave the way you expect.
On WSL you don’t need it.
macOS, Linux and WSL
curl -fsSL https://claude.ai/install.sh | bash
With Homebrew:
brew install --cask claude-code
Homebrew offers two casks. claude-code tracks the stable channel, typically about a week behind, and skips releases with major regressions. claude-code@latest picks up new versions as soon as they ship.
Debian, Fedora, RHEL and Alpine can also install it through apt, dnf and apk.
Verify the Install
claude --version
A version number followed by (Claude Code) means you’re set.
Native installs (the curl and PowerShell route) update themselves in the background.
Homebrew and WinGet don’t. You have to do it yourself:
brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode
Logging In
Type claude in your terminal. The first run brings up the login flow:
claude
Finish authentication in the browser and your credentials are stored, so you won’t be asked again. To switch accounts later, type /login inside the session.
If the ANTHROPIC_API_KEY environment variable is set, the login prompt is skipped and Claude Code asks you to approve that key instead.
Your First Session
Move into your project and start:
cd ~/projects/my-project
claude
Above the prompt you’ll see the version, the active model and the working directory. Make your first input a question, not a command:
what does this project do?
Claude scans the files and summarizes. You don’t paste context, it reads what it needs.
Other useful opening questions:
what technologies does this project use?
where is the main entry point?
explain the folder structure
Run /init to Introduce the Project
Type /init in the session. Claude studies the project and writes a CLAUDE.md guide in the root. That file loads automatically in every new session, so you write your project’s rules once instead of repeating them every time.
Keep it short. A CLAUDE.md past 200 lines eats context in every session. Put in only what the code can’t tell you: build commands, team conventions, traps.
How It Works: Permission Modes
By default Claude Code asks before it changes a file. Shift+Tab cycles through the modes:
| Mode | Behavior | When |
|---|---|---|
| Default | Asks for approval on every change | New projects, unfamiliar code |
acceptEdits | Auto-approves file edits | Routine work where you know what's coming |
plan | Changes nothing, proposes an approach first | Large changes, as the opening move |
Some accounts also have an auto mode. It runs a background safety check and blocks risky actions, returning to prompts only after repeated blocks.
This flag turns off every permission prompt. Claude will use any tool without asking, file deletion included.
Only use it in an isolated, version-controlled environment you can afford to lose. For daily work, plan mode is a far more reasonable starting point.
Your First Code Change
Start small:
add a hello world function to the main file
Claude finds the file, shows you the change, asks for approval and applies it.
Then move to real work. Don’t be vague when reporting a bug:
Weak request:
fix the bug
Strong request:
fix the login bug where users see a blank screen after entering wrong credentials
The difference: the first sentence pushes Claude to scan the whole project, the second takes it straight to the file. Faster result, fewer tokens.
Working with Git
Git turns conversational:
what files have I changed?
commit my changes with a descriptive message
create a new branch called feature/login-screen
help me resolve merge conflicts
Running /diff before you commit is a good habit. One step further is /code-review, which checks the current diff for bugs and cleanup.
8 Habits That Actually Move the Needle
Anyone who installs it can run it. These habits are where the difference shows up.
1. Type /clear when the topic changes. The old context is resent with every message. Once you’ve moved to unrelated work, carrying it costs you and buys you nothing.
2. Open big tasks in plan mode. Shift+Tab into plan mode and read the approach first. If it’s heading the wrong way you fix it before a single line is written.
3. Hit Escape the moment it goes sideways. Don’t wait. Escape stops it instantly; double Escape or /rewind rolls conversation and code back to a checkpoint.
4. Give it something to verify against. “This test should pass”, “output should look like this”, paste a screenshot. When Claude can check its own work, it catches the problem before you have to.
5. Match the model to the job. Sonnet handles most day-to-day coding and costs less. Save Opus for architectural calls and multi-step reasoning. /model switches mid-session.
6. Watch context with /context. It shows what’s taking up room: bloated MCP servers, a long CLAUDE.md, accumulated conversation. The culprit is usually obvious.
7. Ask step by step. Instead of three jobs in one sentence, give a numbered list. Claude works through it in order and you check at each step.
8. Turn off MCP servers you don’t use. Open /mcp and disable the inactive ones. Every server takes a slice of your context.
Claude Code sends the whole conversation with every request. The one-line question you ask at 6pm carries everything that piled up since morning.
Caching lowers that cost but doesn’t erase it. Finished a task? /clear. Taking a break? /rename then /clear. Coming back? /resume.
Updating
Native installs update themselves. To check manually:
claude update
If you suspect something’s wrong with the install, diagnostics are a separate command:
claude doctor
Inside a session, /doctor goes further: duplicate installs, PATH problems, unparseable settings files, slow hooks, and unused skills listed against their context cost.
What’s Next
Once install and the first session are behind you, the next stage is shaping Claude Code around your own work:
- Every command and what it does: the Claude Code slash commands guide
- Writing your own commands, subagents and automation: the skills, subagents and hooks guide
- Connecting external tools: the MCP guide
Frequently Asked Questions
Is Claude Code free? No, there’s no free tier. It runs on a Claude Pro, Max, Team or Enterprise subscription, or on a Claude Console account with prepaid credit. Enterprise users can also connect through Amazon Bedrock, Google Cloud’s Agent Platform and Microsoft Foundry.
How do I install Claude Code on Windows? Open PowerShell and run irm https://claude.ai/install.ps1 | iex. On CMD, use curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd. WinGet works too: winget install Anthropic.ClaudeCode. Confirm with claude --version.
How do I start Claude Code? Open a terminal, cd into your project and type claude. The first run asks you to log in through the browser; after that it opens straight away.
What’s the difference between Claude Code and Claude on the web? The web version gives you the code and leaves the copy-paste to you. Claude Code applies it: finds the file, edits it, runs the tests and shows the result. You don’t paste context either, it reads the project as needed.
Does Claude Code ask before changing my code? In the default mode, yes, it asks for approval on every change. Shift+Tab switches modes: acceptEdits auto-approves edits, plan proposes an approach without changing anything.
How do I update Claude Code? Native installs update in the background. To check manually, run claude update. Homebrew installs need brew upgrade claude-code and WinGet installs need winget upgrade Anthropic.ClaudeCode, since neither auto-updates.
Why does Claude Code use so many tokens? Every request carries the entire conversation. In a session that’s been open for hours, even a one-line question drags along everything that accumulated. Use /clear when the topic changes and /context to see what’s filling the window.
What does CLAUDE.md do in Claude Code? It’s a guide file in your project root that loads automatically in every session. Build commands, team conventions and traps go there, so you don’t explain them each time. The /init command generates it for you. Keeping it under 200 lines is the recommendation.
Summary
The install is one line and the first session takes five minutes. The real difference starts after that.
The three habits that pay the most: /clear when the topic changes, open big tasks in plan mode, and stop writing vague requests. All three are free, and all three move both the result and the bill.
If you came looking for a free version, there isn’t one. But a Console account with a few dollars of credit lets you try it properly, with no monthly commitment 👀