▲ Claude Academy
Sign in

Vault note

Setup Checklist

Everything you need to install or sign up for, once, grouped by what you're actually going to do. Most learners only need the first section. Don't install things you won't use yet — come back to this file when a track tells you to.

Rule of thumb: non-technical and prompting learners need almost nothing. Developers need a working local toolchain. Cloud learners need an AWS or GCP account. Do the section that matches your next track and skip the rest.

(a) Everyone — the account · ⏱ ~5 min

  • Create a claude.ai account (free tier is fine to begin).
  • (Optional) Install the desktop app and mobile app so Claude is always one click away.
  • Bookmark the course home README and your progress checklist.

That's all you need for track-0a-non-technical, track-0b-never-used-claude, and track-1-claude-foundations.

(b) Prompt & evaluation learners — nothing extra · ⏱ 0 min

  • Nothing to install. track-2-effective-prompting runs entirely in claude.ai (and, if you like, the free in-browser prompt tutorials). A spreadsheet app for tracking eval results is nice-to-have but optional.

(c) Developers — local toolchain · ⏱ ~30–45 min

For track-3-building-with-the-api, track-4-claude-code, track-5-mcp-and-tools, and track-6-agents-skills-subagents.

  • Runtime: install Node.js (LTS) and/or Python 3.10+. Node is needed for Claude Code and many MCP examples; Python is great for API and MCP-server work. Having both is ideal.
  • Anthropic API key: create one in the Anthropic Console (console.anthropic.com), add a little credit, and store it as an environment variable so code can read it without hard-coding it: 
    # add to your shell profile (~/.zshrc, ~/.bashrc), then restart your shell
export ANTHROPIC_API_KEY="sk-ant-..."
  • Claude Code: install the CLI and sign in. 
    npm install -g @anthropic-ai/claude-code
claude            # launches; follow the sign-in prompt
  • Editor: a code editor you like (e.g. VS Code), ideally with the Claude Code / terminal integration.
  • git: installed and configured (git --version, set user.name and user.email). You'll version every project.
  • Quick smoke test: run a one-line API call (you'll do this for real in p05-first-api-call) and open Claude Code in any folder to confirm both work.

(d) Cloud deployment — pick your platform · ⏱ ~30 min

Only for track-7-deployment-and-scale. You need access to one of these (your employer may already provide it):

  • AWS / Amazon Bedrock: an AWS account with Bedrock access, Claude models enabled in your region, and AWS credentials configured locally (aws configure). Backing course: claude-with-amazon-bedrock.
  • Google Cloud / Vertex AI: a GCP project with the Vertex AI API enabled, Claude models granted in Model Garden, and gcloud auth set up (gcloud auth application-default login). Backing course: claude-with-vertex-ai.

⚠️ Security note — API keys are secrets

  • Never commit an API key to git, paste it into a chat, or share it in a screenshot. Treat it like a password.
  • Keep keys in environment variables or a secrets manager, never hard-coded in source. Add .env to your .gitignore.
  • If a key is ever exposed, rotate (revoke and regenerate) it immediately in the Console.
  • Use the least privilege that works: separate keys per project, set spend limits, and delete keys you no longer use.
  • For cloud, prefer the platform's IAM roles/service accounts over long-lived static credentials where you can.

See also