How to Start with Claude Code (2026): From Install to First Shipped Feature

1. Install the CLI (2 min)

The fastest route is the native installer:

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

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

# or via npm, if you prefer
npm install -g @anthropic-ai/claude-code

Verify with claude --version. It runs on macOS, Linux and Windows; there are also VS Code and JetBrains extensions plus a desktop app, but learn the terminal first — everything else is the same engine.

2. Authenticate (1 min)

Run claude in any project folder. On first launch it opens a browser to log in. Two ways to pay:

• Subscription (recommended): Claude Pro (~$20/mo) covers casual use; Max plans (~$100–200/mo) cover serious daily agent work. Flat cost, no surprises.
• API key: pay-as-you-go per token. Fine for occasional use; for daily agentic coding it usually costs more than Max — check our token cost calculator.

3. Your first session: just talk to your repo (5 min)

Open a real project and start with questions, not changes:

cd your-project
claude

> what does this codebase do? give me the 5 most important files
> how does authentication work here?
> find every place we log user emails

This is the fastest way to build trust in what it can see. It reads files on demand — nothing is uploaded ahead of time. When you're ready for a first change, keep it small and verifiable:

> add input validation to the signup form, then run the tests

Claude Code will propose file edits and ask permission before running commands. Approve each step until you trust it, then use auto-accept mode (shift+tab) for routine work.

4. Create CLAUDE.md — the step everyone skips (3 min)

CLAUDE.md is a file in your repo root that gets loaded into every session: build commands, conventions, gotchas. It's the difference between a generic assistant and one that knows your project. Generate a starter automatically:

> /init

Then edit it down to what matters. Ours look like:

# CLAUDE.md
- pnpm monorepo: build with `pnpm build`, test with `pnpm test`
- NEVER edit generated files in /dist
- DB changes: additive Prisma migrations only
- code style: match neighboring files, no extra comments

5. Ship a real task (5 min)

The workflow that works for feature-sized tasks:

• Describe the outcome, not the steps: "Add a /health endpoint that checks DB connectivity, with a test. Follow the pattern in routes/status.ts."
• Ask for a plan first on bigger tasks: "make a plan before writing code" — you catch misunderstandings before they cost tokens.
• Let it verify itself: end tasks with "run the tests and fix any failures". Agents shine when they can check their own work.
• Review the diff, then: "commit this with a sensible message" — it writes clean commits (and PRs, if you use GitHub).

Where to go next

• MCP (Model Context Protocol): connect Claude Code to your database, browser or internal APIs — claude mcp add. This is where it graduates from "edits files" to "operates your stack".
• Custom slash commands & skills: reusable prompts for your team's repetitive workflows.
• Deciding between tools? Read Claude Code vs Cursor — the honest version of the question everyone asks.