Getting Started with Claude Code: Install to First PR

Claude Code is Anthropic’s agentic coding tool. Unlike a chat window, it works inside your project: it reads files, runs commands, edits code and works through problems while you watch or redirect. This guide takes you from zero to a productive first session, with every command checked against the official docs as of October 2026.

Before you begin

You need:

• a terminal and a code project;
• one of: a Claude Pro, Max, Team or Enterprise subscription, a Claude Console account (API credits), or access through Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry.

System requirements: macOS 13.0+, Windows 10 1809+ (or Server 2019+), Ubuntu 20.04+, Debian 10+ or Alpine 3.19+, with 4 GB+ RAM and an internet connection.

Not a terminal person? Claude Code also runs as a desktop app, in VS Code and JetBrains IDEs, and on the web at claude.ai/code. The concepts below apply everywhere.

Step 1: Install Claude Code

Native installer (recommended, auto-updates):

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell
irm https://claude.ai/install.ps1 | iex

Package managers (no auto-update):

brew install --cask claude-code        # macOS (stable channel)
winget install Anthropic.ClaudeCode    # Windows

npm (needs Node.js 22+):

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

Don’t use sudo npm install -g. Open a new terminal and confirm:

claude --version

If the shell says claude isn’t found, the install directory isn’t on your PATH yet; our Claude Code install guide covers PATH fixes, Windows options, updates and uninstalling. On native Windows, installing Git for Windows lets Claude Code use its Bash tool; otherwise it uses PowerShell.

Step 2: Start your first session and sign in

cd /path/to/your/project
claude

On first run, Claude Code prompts you to log in through your browser. Credentials are stored, so you only do this once. Use /login later to switch accounts.

The prompt shows the version, current model and working directory. Type /help to see available commands.

Step 3: Ask about your codebase

Start read-only. Claude reads files as needed; you don’t have to paste them.

what does this project do?

where is the main entry point, and how does a request flow through the app?

explain the folder structure

This is also a good way to onboard onto an unfamiliar repo.

Step 4: Make your first change

add input validation to the user registration form

Claude finds the right files, proposes the edit and, depending on your permission mode, asks before applying it. Choose Yes to approve. For one feature taken all the way from question to pull request, follow our Claude Code tutorial.

Permission modes

Press Shift+Tab to cycle modes. The status bar shows the active one. From v2.1.283, interactive terminal and VS Code sessions start in Auto mode when nothing else sets a mode and auto mode is available; claude -p runs start in Manual.

Mode	Config value	What Claude does without asking
Manual	default	Reads only; asks before edits, commands and network
Accept edits	acceptEdits	Edits files without asking
Plan	plan	Researches and proposes a plan; no edits
Auto	auto	Uses a safety classifier to approve or deny actions (where available)

There is also bypassPermissions, which skips prompts entirely; reserve it for isolated environments. Fine-grained allow/ask/deny rules live in /permissions; the permission modes guide covers every mode, including auto mode.

Step 5: Use git conversationally

what files have I changed?

commit my changes with a descriptive message

create a new branch called feature/registration-validation

Step 6: Create a CLAUDE.md

CLAUDE.md is the file Claude reads at the start of every session in that project (our CLAUDE.md guide has templates for five stacks). Generate a starter:

/init

Claude analyses the codebase and writes build commands, test instructions and conventions it discovers. If a CLAUDE.md exists, /init suggests improvements instead of overwriting.

Then edit it down. A good CLAUDE.md:

• lists commands Claude can’t guess (pnpm test:unit, make dev-db);
• states conventions that differ from defaults (“use Bun, not npm”);
• notes architecture decisions and gotchas;
• stays under about 200 lines, the official target, because longer files reduce adherence.

# Project notes

## Commands
- Dev server: `pnpm dev`
- Unit tests: `pnpm test:unit` (run before every commit)
- Typecheck: `pnpm typecheck`

## Conventions
- TypeScript strict mode; no `any`
- API handlers live in `src/server/routes/`, one file per resource
- Use the `logger` helper, never `console.log`

## Gotchas
- The `legacy/` folder is frozen; don't edit it

Where the different memory files go:

File	Scope
./CLAUDE.md	Project, shared with the team via git
./CLAUDE.local.md	Project, personal (add to .gitignore)
~/.claude/CLAUDE.md	All your projects
.claude/rules/*.md	Topic-specific rules, optionally scoped to file paths

You can import other files with @path, for example See @README for an overview. Run /memory to edit these files and manage auto memory.

Step 7: Learn the essential commands

Shell commands

Command	What it does
claude	Start an interactive session
claude "task"	Start with an initial prompt
claude -p "query"	Run once non-interactively and exit (scripts, CI)
claude -c	Continue the most recent conversation in this directory
claude -r	Resume a previous conversation from a list

Slash commands

Command	What it does
/help	Show available commands
/init	Create or improve CLAUDE.md
/clear	Start a new conversation with empty context
/compact [instructions]	Summarise the conversation to free context
/context	Visualise what’s using your context window
/model	Switch model (and effort level, where supported)
/plan [task]	Enter plan mode
/rewind	Restore conversation and/or code to a checkpoint
/resume	Pick up an earlier session
/permissions	Manage allow, ask and deny rules
/mcp	Manage MCP server connections and sign-in
/hooks	View configured hooks
/plugin	Browse and install plugins
/usage	Session cost, plan limits and activity (/cost is an alias)
/doctor	Diagnose setup issues and propose fixes
/code-review	Review the current diff (alias /review)
/btw [question]	Ask a side question without adding it to history

Type / at any time to see everything available, including your own skills. For the full list, see our Claude Code slash commands cheatsheet.

Step 8: Work the way experienced users do

These habits come straight from Anthropic’s Claude Code best-practices guide; our Claude Code best practices page covers 20 of them in depth.

Give Claude a way to verify its work

The biggest single improvement. Instead of “implement email validation”, say: “write a validateEmail function; user@example.com is valid, user@.com is not; write tests and run them”. With a pass/fail signal, Claude iterates until the check passes rather than stopping when the work merely looks done.

Explore, plan, implement, commit

1. Explore in plan mode (Shift+Tab until “plan mode on”, or claude --permission-mode plan): “read /src/auth and explain how sessions work”.
2. Plan: “I want to add Google OAuth. What files change? Create a plan.” Press Ctrl+G to edit the plan in your editor.
3. Implement: approve the plan and let Claude code and run tests.
4. Commit: “commit with a descriptive message and open a PR”.

Skip the plan for changes you could describe in one sentence.

Be specific

“Fix the login bug where users see a blank screen after entering wrong credentials” beats “fix the bug”. Point at files, paste error messages, include screenshots for UI work.

Manage context

• /clear between unrelated tasks.
• /compact Focus on the API changes to summarise with a focus. The context management guide explains when to use each.
• Delegate investigations to subagents: “use subagents to investigate how token refresh works”. See the subagents guide.
• If you’ve corrected Claude twice on the same issue, /clear and restart with a better prompt.

Course-correct early

Press Esc to stop Claude mid-action (context is kept). Press Esc twice or run /rewind to roll back. Checkpoints only cover edits made through Claude’s file tools, not Bash side effects, so git is still your safety net.

Step 9: Extend Claude Code

Once the basics feel natural, build your Claude kit:

• Skills for procedures you repeat.
• Subagents for isolated, specialised work.
• Hooks for things that must happen every time.
• MCP servers to connect GitHub, docs, browsers and databases.
• Plugins to install or share bundles of the above.

Track spend with /usage (our Claude Code pricing guide explains plans and limits), and analyse usage over time with ccusage. Wondering how it stacks up against an AI editor? Read Claude Code vs Cursor. And sharpen your requests with our Claude prompting best practices.

Troubleshooting quick reference

• claude: command not found: open a new terminal; check your PATH.
• Install fails with a 403 or HTML error: see the official install troubleshooting page for alternatives.
• Something feels off: run /doctor in a session, or claude doctor in your shell for read-only diagnostics. More answers in the Claude Code FAQ.
• Claude ignores CLAUDE.md rules: shorten the file, remove contradictions, or convert must-follow rules into hooks.