Claude Code for Everyone

Module 07 — Claude Code Setup

🎯 Goal: Install Claude Code, sign in, and have your first real conversation with it. ⏱️ Time: ~1 hour.

What you're installing

Claude Code is a command-line program. You start it by typing claude in your terminal, inside whatever project folder you want it to work in. Then you chat.

Step 1 — Install Node.js (a prerequisite)

Claude Code runs on Node.js, so install that first with Homebrew:

brew install node

Confirm:

node --version

(You won't use Node directly — it's just the engine Claude Code runs on.)

Step 2 — Install Claude Code

npm install -g @anthropic-ai/claude-code
  • npm is Node's installer (came with Node).
  • -g means "globally" — available from any folder.

Confirm:

claude --version

If npm install -g gives permission errors on your corporate laptop, there's also a native installer — see Troubleshooting. Don't use sudo to force it; that causes worse problems later.

Step 3 — Sign in

You need a Claude account with a plan that includes Claude Code. Plans and pricing change over time, so check the current options on Claude's plan page — as a rough guide, the higher tiers give more usage for heavy automation, and a lighter plan is fine to start. (A technical friend can help pick — this is the one thing that costs money.)

  1. In Ghostty, type:

    claude

  2. The first time, it walks you through signing in — it'll open a browser window to log in to your Anthropic account. Follow the prompts.

  3. Once signed in, you'll see the Claude Code prompt waiting for your message.

Step 4 — Your first conversation (the fun part)

Always run claude inside a project folder, so it has a place to work. Let's use the practice project from Module 06:

cd ~/my-first-project
claude

Now just talk to it. Try these, one at a time, reading each response:

1. Get oriented:

What files are in this project and what does each one do?

2. Make it do something:

Create a Python program called greet.py that asks for my name and then prints a friendly hello with today's date. Then run it to show me it works.

Watch it: it writes the file, you see it appear in VS Code's sidebar, it runs it, and shows you the output. That's the whole magic. You described; it built.

3. Iterate — the real skill:

Now also make it tell me what day of the week today is.

Notice you didn't re-explain everything. Claude Code remembers the conversation. You refine by talking, like with a colleague.

How to talk to Claude Code (the most important skill in this course)

The tools were easy. This is the actual skill. A few principles:

  1. Be specific about the outcome, not the method. You don't say "use a for loop" — you say "go through every PDF in this folder." Describe the what; it figures out the how.

  2. Give it the context it needs. "Pull out the client name" is vague. "Pull out the client name, which always appears after the label 'Member:' near the top" is gold. You know your documents; tell it what you know.

  3. One step at a time for big jobs. Instead of one giant request, build up: "First, just list all the PDFs and tell me how many there are." Then: "Good, now for the first one, show me what text you can read from it." Then expand. This catches misunderstandings early.

  4. When something's wrong, say what you expected vs. what happened. "It put the date in the wrong column — the date should be column B, not column D." Precise feedback gets precise fixes.

  5. Ask it to explain. "Why did you do it that way?" or "explain that last step simply." It's a tutor as much as a tool.

  6. Ask before destructive actions. If it proposes deleting or overwriting files, you can always say "show me what you'll delete first." It will.

Two power features to know about (don't over-use yet)

  • CLAUDE.md file: If you create a file named CLAUDE.md in a project, Claude Code reads it every time as standing instructions. Great for "always save outputs to the results folder" or "client files use the format LASTNAME_FIRSTNAME." You can even ask Claude Code to create and maintain it: "set up a CLAUDE.md describing how I like this project organized."
  • Plan mode: For bigger tasks, you can ask it to plan first before doing anything: "make a plan for this, don't change anything yet." You review the plan, then say go. Safer for important work.

Safety reminders specific to Claude Code

  • It will ask permission before running commands that change things — read those prompts; don't just approve blindly. When unsure, ask it "what does this do?"
  • Keep the data rule in mind: working on files locally is fine; don't have it send file contents to external services.
  • git commit your work before big changes (Module 05). Claude Code can even do this for you: "commit everything first, then make the change."

✅ You're done with this module when

  • claude --version works and you've signed in.
  • You had a back-and-forth conversation that created and ran a small program.
  • You successfully refined something by asking for a change.
  • You understand the 6 principles of talking to Claude Code (especially: describe the outcome, give context, go step by step).

If a box isn't checked yet: see Troubleshooting → Claude Code won't install, or if install is blocked, the native installer noted there avoids npm. Sign-in trouble is usually a plan issue — check Claude's current plan page.

Next: Module 08 — Your First Automation.