How to Set Up Claude Code on Your Mac

If you've used Claude (or ChatGPT) in a browser tab, you've used about 30% of what these tools can do.

The browser version is locked in the tab. It can't open your files. It can't run a command. It can't update your Notion. It can talk to you, and that's it.

The CLI version — Claude Code — is the same intelligence with the lid off. It can read your files, run commands, edit code, draft an email, update a Notion page, search your calendar, deploy a site. Anything you can connect, it can use.

This is the foundation. Set this up once and everything else — the agent persona, the Telegram bot, the automations — is built on top.

Who this is for

You're comfortable using your computer day-to-day. You can open Terminal once, paste a command, and follow on-screen prompts. You're not a developer (and you don't need to be).

If you can install an app from a website, you can do this.

What you'll need

• A Mac (Apple Silicon or Intel — same steps either way)
• 30 minutes
• A Claude Pro or Max account (Claude Code isn't on the free tier, and you'll need Pro anyway to add the connectors that make it useful)

---

Step 1: Open Terminal

It's already on your Mac. Hit Cmd+Space, type "Terminal," press Enter.

A small window with a blinking cursor opens. Don't be intimidated. You'll paste two commands total.

Step 2: Install Claude Code

Paste this and hit Enter:

curl -fsSL https://claude.ai/install.sh | bash

Wait for it to finish — it'll download Claude Code and drop it in ~/.local/bin/claude.

When it's done, close Terminal, open a fresh window, and type:

claude

If you see a welcome screen: you're in.

If you see "command not found," restart Terminal and try once more. Still broken? Paste this into Claude on the web and follow its instructions:

I installed Claude Code via the curl install.sh script on my Mac. The claude command says "command not found" in a new Terminal window. Walk me through diagnosing and fixing it — check $PATH, verify ~/.local/bin/claude exists, and tell me what to add to my shell config (.zshrc or .bash_profile) so the command is permanently available.

Step 3: Sign in

The first time you run claude, it walks you through signing in. Use the same account you log into claude.ai with.

You'll land at a > prompt waiting for input. That's your AI's chat window. Type a question, hit Enter.

Try: "What's in my home directory?"

It runs ls, sees what's there, and tells you. That's the difference. It has hands.

Step 4: Give it superpowers (MCPs)

Without MCPs, Claude Code can read and edit files on your computer. Useful, but not the full picture.

MCPs are connectors that let Claude reach into your other apps: Gmail, Calendar, Drive, Notion, Slack, your CRM, your project tool. Each MCP gives Claude one new ability inside an app you already use.

Set them up at claude.ai under Settings → Connectors. Anything you connect there shows up in Claude Code on your Mac automatically. No extra setup.

Start with the obvious ones:

• Gmail — read, search, draft
• Google Calendar — read events, schedule things
• Google Drive — search and read your files
• Notion — read and write pages

Add more as you go. The more connected, the more your AI can actually do without you doing the busywork.

Step 5: Teach it about you (CLAUDE.md)

By default, Claude knows nothing about you. Every conversation starts cold.

Fix this with a file called CLAUDE.md — a plain Markdown file Claude reads at the start of every session in that folder. Put it in the folder where you'll work most often. (Mine lives in my Obsidian vault, which is where my whole second brain sits.)

What to put in it:

• Your name and role
• The tools you use
• How you like Claude to behave (short answers? Don't ask permission for X? Always confirm before sending email?)
• Anything you don't want to re-explain every session

Don't overthink it. A 20-line CLAUDE.md is plenty. You'll add to it as you discover what you want.

Paste this prompt to bootstrap one:

I'm setting up Claude Code for the first time. Help me draft a CLAUDE.md file for the folder I'll use as my "home base." Ask me 5-7 quick questions about who I am, what I do, what tools I use, and how I want you to behave. Then write the CLAUDE.md and save it to this folder.

Step 6: Set permissions (.claude/settings.json)

By default Claude asks before every action — read a file, edit a file, send an email. After ten minutes of "are you sure?" you'll want to tell it which actions it can take without asking.

In the same folder as your CLAUDE.md, create a .claude/settings.json file that defines an allowlist.

The pattern: read = auto-approve. Write to safe places = auto-approve. Anything outbound or destructive = ask first.

Paste this prompt:

Set up a .claude/settings.json in this folder. Auto-approve: Read, Glob, Grep, LS, and Edit/Write within this folder. Auto-approve any MCP tool whose name starts with list, get, search, fetch, or read. Ask before any tool that sends, creates, updates, or deletes — especially anything that touches Gmail, calendar invites, Notion writes, or money. Show me the file and explain each section.

You can revise as you go. Most people loosen things up over time once they trust what their setup will and won't do on its own.

Step 7: (Optional) Skills for things you do often

If you find yourself giving Claude the same instructions over and over — "draft a LinkedIn post in my voice using this framework," "summarize this meeting into action items the way I like" — you can save those as Skills.

A Skill is a folder with a SKILL.md (the instructions) and any helper files. Invoke it by typing /skill-name in Claude Code.

Build them later, once you've found the patterns. Don't build skills for things you'll only do once.

What to do next

You now have:

• Claude Code installed
• Connected to your apps via MCPs
• A CLAUDE.md teaching it about you
• A settings.json defining what it can do on its own

From here:

1. Use it. Open Terminal, type claude, and ask it to do something real. A weekly recap. A draft. A search across your inbox.
2. Add MCPs as you discover what you wish it could touch.
3. When you're ready to text your AI from anywhere, the next step is the Telegram bridge.

If you'd rather have one built for you

This is exactly what I set up for executives — your apps, your voice, your guardrails — in one focused session. If you want help, that's the One Hour AI Setup. Show up, leave with a working system.