Quickstart

0. Which surface should I use?

1. claude.ai (web)

1. Open claude.ai.
2. Sign up with Google, Apple, or email + password.
3. Click your avatar → Settings and turn on:
   • Artifacts — live previews for code, diagrams, and small apps.
   • Skills — the four document skills (PDF, Word, Excel, PowerPoint) are toggled here. Custom skills can be uploaded as folders.
   • Connectors — Google Drive, Calendar, GitHub, Notion, Figma, and more. Each shows up as MCP tools Claude can call mid-conversation.
   • Memory (Pro+) — persistent memory across conversations.
   • Custom instructions — your "always do X" preferences.
4. Switch the model picker to Claude Opus 4.7 for hard tasks, Sonnet 4.6 for daily work, Haiku 4.5 for speed/cost.

Free vs Pro vs Max

2. Desktop app (macOS / Windows)

Same account, same skills, same settings as claude.ai — but with a system-wide shortcut and proper file-drag-and-drop.

1. Download from claude.ai/download.
2. Sign in with the same account as the web app.
3. Set the global shortcut (default ⌘ ⌥ Space on macOS, Ctrl Alt Space on Windows) so Claude is always one keystroke away.
4. Drag files directly into the chat window — works for PDFs, images, code, anything Claude can read.

3. Claude Code CLI

This is what pros use for actual coding work. It's an interactive terminal agent that reads your repo, edits files, runs commands, opens PRs.

Install

# npm (cross-platform)
npm install -g @anthropic-ai/claude-code

# or native installer (macOS / Windows / Linux)
# https://docs.claude.com/en/docs/claude-code/install

First run

# in a project directory
cd ~/code/my-project
claude

On first run it prompts you to authenticate. Two paths:

• Subscription — log in with your claude.ai Pro or Max account. Usage counts against your subscription limits.
• API key — pay per token, generated at console.anthropic.com. Best for heavy users or team deployments.

Bootstrap the project for Claude

# inside Claude Code, on day one
/init

This scans the repo and writes a CLAUDE.md file with architecture summary, commands, and conventions. Every future session reads this on startup. See section 8.

The five things to configure right away

1. Permissions — ~/.claude/settings.json → which Bash commands auto-allow. Reduces interruptions. See the config section.
2. Skills — drop folders into ~/.claude/skills/. Start with the official document and Anthropic skills.
3. MCP servers — connect GitHub, Postgres, Linear, etc. See the MCP guide.
4. Slash commands — drop reusable prompt templates into ~/.claude/commands/. See commands.
5. Hooks — automate things on tool use, stop, session start. See hooks.

Day-to-day flow

# open in your project
claude

# ask in plain English; it'll plan, read, edit, run tests
> Add a /healthz endpoint with a JSON body that includes the git SHA.

# useful built-ins
/init             # write or update CLAUDE.md
/review           # review the current branch's diff
/security-review  # OWASP-style sweep of the diff
/skills list      # see what skills are loaded
/mcp list         # see connected MCP servers
/help             # everything else

Tips that turn beginners into pros

• Write a CLAUDE.md early. The 10 minutes you spend up front saves hours.
• Use /init in every new repo so future sessions have context.
• Trust the agent on small tasks, supervise on big ones. The /plan mode is your friend for anything risky.
• Skill > prompt. If you find yourself re-pasting the same instructions, make it a skill or slash command.
• Permissions are friction reducers. Allow safe read-only Bash commands once, never again.

4. IDE extensions

Claude Code ships an IDE extension that gives you the same agent inline. Available for VS Code and the full JetBrains family (IntelliJ, PyCharm, GoLand, WebStorm, Rider).

VS Code

1. Extensions panel → search "Claude Code" → install.
2. Open the command palette (⌘ ⇧ P) and run Claude: Sign In.
3. Choose subscription or API key.
4. Open the Claude side panel (default shortcut ⌘ ⇧ C) to chat with your project.

JetBrains

1. Settings → Plugins → Marketplace → search "Claude Code" → install → restart IDE.
2. Sign in via the Claude side tool window.
3. Right-click in the editor for "Ask Claude about this" / "Refactor with Claude".

5. The Anthropic API

For automations, your own apps, and anything that needs to run unattended. Pay-per-token via console.anthropic.com.

Get an API key

1. Go to console.anthropic.com → sign up or sign in.
2. Settings → API Keys → Create Key.
3. Add billing under Settings → Plans & Billing. Add a usage limit so a runaway script doesn't drain your card.
4. Copy the key to a password manager. Never commit it to git.

First call (Python)

pip install anthropic
# put your key in env
export ANTHROPIC_API_KEY=sk-ant-...

from anthropic import Anthropic
client = Anthropic()

resp = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Say hi in haiku."}],
)
print(resp.content[0].text)

First call (Node)

npm install @anthropic-ai/sdk

import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic();

const resp = await client.messages.create({
  model: "claude-opus-4-7",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Say hi in haiku." }],
});
console.log(resp.content[0].text);

That's the smallest possible call. For everything else — prompt caching, extended thinking, tool use, batch, vision, files, citations — see the full API guide.

6. Mobile apps

iOS and Android apps mirror claude.ai with one big addition: voice mode. Hold the mic icon, talk, and Claude speaks back. Camera input also works — point at a problem, get an answer.

• iOS — App Store
• Android — Play Store

7. First-day configuration (Claude Code)

Open ~/.claude/settings.json and add the things every pro user adds in their first week:

{
  "permissions": {
    "allow": [
      "Bash(git status)",
      "Bash(git diff:*)",
      "Bash(git log:*)",
      "Bash(npm test)",
      "Bash(npm run lint)",
      "Bash(pytest:*)",
      "Bash(go test:*)",
      "Read(/Users/me/code/**)"
    ],
    "deny": [
      "Bash(rm -rf:*)",
      "Bash(git push --force:*)"
    ]
  },
  "env": {
    "EDITOR": "code --wait"
  },
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [{ "type": "command", "command": "prettier --write \"$CLAUDE_FILE\"" }]
      }
    ]
  }
}

What this does:

• Auto-allows safe read-only Bash calls — no more permission prompts mid-flow.
• Hard-denies destructive commands so a mistake stays a prompt, not an apocalypse.
• Sets EDITOR so Claude opens the right editor for interactive prompts.
• Auto-formats any file Claude edits with Prettier (replace with your formatter — black, gofmt, rustfmt).

See Hooks for the full event list (PreToolUse, PostToolUse, Stop, Notification, UserPromptSubmit, SessionStart, SubagentStop) and example automations.

8. Writing your first CLAUDE.md

Run /init in a fresh project and Claude writes one for you. Then edit it — the magic happens when you add the things only you know.

What goes in

• Architecture (1 paragraph) — the shape of the codebase.
• Commands — how to install, run, test, lint, deploy.
• Conventions — naming, error handling, where tests live, lint rules.
• Gotchas — flaky tests, side-effecting files, deprecated paths to avoid.
• "Always do X" rules — e.g., "always run npm run typecheck after edits".

A real-world template

# CLAUDE.md

## Project

Server-rendered Next.js app + Postgres. ~30k LOC TypeScript.
Deploys to Vercel on merge to main.

## Run / test / lint

- Install:    pnpm install
- Dev:        pnpm dev             (port 3000)
- Typecheck:  pnpm tsc --noEmit
- Test:       pnpm test            (Vitest)
- Lint:       pnpm lint            (Biome)
- Migrations: pnpm db migrate

## Conventions

- Routes in app/, server actions in app/_actions/.
- Components colocated; shared in components/.
- DB access via db/queries/; never inline.
- Test naming: *.test.ts next to the file.

## Always

- Run pnpm tsc --noEmit after any edit.
- Commit messages: Conventional (feat/fix/refactor).
- New env vars: also add to .env.example.

## Gotchas

- lib/legacy/ is deprecated; do not extend.
- tests/e2e/ is flaky on CI; if it fails twice, rerun.

9. Where to go next

• Skills directory — 101 skills to install today.
• MCP guide — connect external tools (GitHub, Postgres, Linear, Figma, Slack).
• Plugins — the marketplace and what's worth installing.
• Slash commands — built-ins + how to write your own.
• Hooks — automate format/test/notify around tool use.
• API guide — prompt caching, extended thinking, tool use, batch.
• Models — Opus 4.7 vs Sonnet 4.6 vs Haiku 4.5 — when to use which.