Zero To Claude Operator In A Weekend

The complete non-technical setup for Claude Code: install commands, the exact CLAUDE.md template, the 4 skills worth installing, 6 MCP connectors with GitHub links, the prompting patterns to use daily, and a 12-minute walkthrough to ship your first app to Vercel by Monday morning. Eight steps, no code, two evenings of focused work.

What You'll Learn

• The verified install commands for Claude Code plus auth in 5 minutes
• The exact CLAUDE.md template you can copy line by line
• The top 4 generic-beginner skills (with GitHub repos and install commands)
• The /skill-creator pattern that turns any repeating prompt into a saved skill
• 6 MCP connectors with install links: Drive, Notion, GitHub, HubSpot, Slack, Gmail
• The talk-like-an-employee prompting pattern plus example prompts
• A 12-minute deploy walkthrough that ships a working app to Vercel

The order matters. Most people skip step 2 (CLAUDE.md) and wonder why Claude keeps misunderstanding them. Most people skip step 5 (MCPs) and wonder why Claude can't read their files. Run the steps in order the first time.

Step 1: Install Claude Code

Claude Code is a CLI tool you install once. It runs in your terminal but you talk to it in plain English. You don't need to know how to code. The terminal is just the most powerful place for Claude to read and write files, run scripts, and connect to your tools. Two commands and you're in.

Prerequisites:

• macOS, Windows, or Linux
• Node.js v18 or higher installed (download: nodejs.org/en/download)
• An Anthropic account at claude.com (free tier works to start)
• A terminal app (Terminal on macOS, PowerShell on Windows, any shell on Linux)

Install:

npm install -g @anthropic-ai/claude-code
claude

The first command installs the Claude Code CLI globally. The second starts a session and prompts you to authenticate the first time.

Authentication: On first launch Claude Code opens a browser tab asking you to sign in with your claude.com account. Sign in, approve access, done. Your session stays authenticated until you log out.

Bookmarks: Official docs at docs.claude.com/en/docs/claude-code/overview. Official landing page with installers at claude.com/claude-code.

Step 2: Configure Your CLAUDE.md File

CLAUDE.md is a markdown file that lives at ~/.claude/CLAUDE.md and gets loaded into every single Claude Code session. It tells Claude who you are, how you write, what tools you use, and how you want it to behave by default. Skip this step and Claude treats you like a stranger every session. Do it and Claude treats you like it's worked with you for years.

Create the file (macOS / Linux):

mkdir -p ~/.claude && touch ~/.claude/CLAUDE.md
open ~/.claude/CLAUDE.md

Or in Claude Code itself, just say: "Open my CLAUDE.md file for me"

Copy this template (edit for yourself):

# Who I am
Your Name. Your role. Your audience.
Example: "Solo builder. Non-engineer.
Audience: builders shipping personal software."

# How I write
Your tone rules. Things to avoid.
Example: "Conversational, direct.
Avoid corporate-speak. First-person."

# Tools I use
Your stack.
Example: "Claude Desktop, Claude Code, MCPs for
Drive + Notion + HubSpot + Gmail."

# Default behavior
Ground rules for Claude.
Example:
- never overwrite my work without asking
- always show your plan first
- end with a clear next step

Karpathy's 4 rules: Andrej Karpathy wrote the most-shared CLAUDE.md guide on GitHub. His 4 rules: (1) keep it under 200 lines, (2) write it in markdown headers Claude can scan, (3) include 1 or 2 short examples of how you want responses formatted, (4) update it weekly based on what Claude got wrong that week. Reference: github.com/karpathy/nanochat

Step 3: Install The Top 4 Skills

Skills are reusable prompts packaged as a single command. Type /skill-name and Claude runs the saved workflow. Install these 4 first. They cover 80% of what you'll need before you build your own.

Skill 1, /skill-creator. The meta-skill. Lets you build your own skills in 3 minutes by describing what you want in plain English. The most important skill you can install because it lets you stop installing skills.

/plugin install skill-creator@anthropic-skills

Source: github.com/anthropics/skills (Anthropic's official skill repo)

Skill 2, /humanizer. Paste any AI-generated text and /humanizer strips the giveaway phrases ("delve into", "in conclusion", the rule of three, dash overuse, etc). Output reads like a human wrote it. Anyone publishing text online needs this.

/plugin install humanizer@coopersimson
/humanizer <paste your text>

Source: github.com/coopersimson96/ai-content-system

Skill 3, /research. Deep web research with cited sources in one prompt. Claude pulls from multiple search engines, scrapes the top results, and returns a synthesized brief with footnoted sources. Saves hours of tab-switching.

/plugin install research@anthropic-skills
/research "top 5 hubspot mcp servers in 2026"

Source: github.com/anthropics/skills/tree/main/research

Skill 4, /awesome-claude-skills. A curated, community-maintained index of every Claude skill worth installing, sorted by category (productivity, coding, content, research). Treat it like a marketplace browser.

/plugin install awesome-claude-skills@hesreallyhim
/awesome-claude-skills

Source: github.com/hesreallyhim/awesome-claude-code (star it and check monthly)

Step 4: Build Your Own Skill

Anytime you find yourself typing the same kind of prompt twice, turn it into a skill.

1. Notice the repeat. You catch yourself asking Claude the same kind of thing twice. "Draft a tweet for me about X." "Draft a tweet for me about Y." Same shape, different topic. That's a skill.
2. Run /skill-creator. Inside any Claude Code session, type /skill-creator and describe the pattern: "I want a skill called /draft-tweet. It takes a topic, writes 3 hook variants in my voice, and adds a comment for which one I should pick."
3. Save and reuse. Claude writes the skill markdown, asks you to confirm, then saves it to ~/.claude/skills/draft-tweet/. From now on, type /draft-tweet with a topic and you get exactly what you asked for. Forever.

Example /draft-tweet skill spec (the file starts with a standard frontmatter block giving the skill's name and description, then the instructions):

name: draft-tweet
description: 3 hook variants of a tweet in my voice + a pick

Given a TOPIC the user provides:
1. Write 3 distinct hook variants (under 280 chars each).
   - Variant 1: contrarian take
   - Variant 2: personal confession
   - Variant 3: data point or stat hook
2. After the 3 variants, write one line:
   "PICK: variant #N because <one-sentence reason>."
Voice rules: lowercase, direct, first-person.

One skill = 100 prompts you never type again. Most people use Claude as a chat assistant. Operators use Claude as a skill library. The shift from asking to invoking is the single biggest unlock in this whole guide. Build 5 skills the first weekend.

Step 5: Connect Your Apps Via MCPs

MCPs (Model Context Protocol servers) are the bridge between Claude and your other tools. Once an MCP is installed, Claude can read AND write to that app: read your last 20 emails, create a Notion page, pull a Google Doc, update a HubSpot deal. These 6 cover 90% of the apps a non-engineer touches daily.

MCP 1, Google Drive. Claude reads and writes Docs, Sheets, Slides. Easiest install: claude.ai Settings, then Connectors, then Google Drive, then Connect. Or install the official server:

npm install -g @modelcontextprotocol/server-gdrive

Source: github.com/modelcontextprotocol/servers/tree/main/src/gdrive

MCP 2, Notion. Read pages and databases, create new pages with full block content, update properties. Available directly in claude.ai Settings, then Connectors, then Notion. Or:

npm install -g @makenotion/notion-mcp-server

Source: github.com/makenotion/notion-mcp-server

MCP 3, GitHub. Read repo files, search code, create issues, open PRs. Essential if you're going to vibe-code anything. Connect via claude.ai Settings, then Connectors, then GitHub (authorize via OAuth). Or:

npm install -g @modelcontextprotocol/server-github

Source: github.com/github/github-mcp-server

MCP 4, Gmail. Read your inbox, draft replies, search threads, manage labels. Connect via claude.ai Settings, then Connectors, then Gmail. Or:

npm install -g @modelcontextprotocol/server-gmail

Source: github.com/GongRzhe/Gmail-MCP-Server

MCP 5, HubSpot. Read contacts, update deals, log activities, search the CRM. Massive unlock for solo founders running outbound or sales.

npm install -g @modelcontextprotocol/server-hubspot

Or use HubSpot's beta direct integration: developers.hubspot.com/docs/mcp. The Pipedream MCP index (github.com/PipedreamHQ/awesome-mcp-servers) covers HubSpot plus 1000+ other apps.

MCP 6, Slack. Read channel messages, post replies, search threads. Auth: create a Slack app with a bot token at api.slack.com/apps.

npm install -g @modelcontextprotocol/server-slack

Source: github.com/modelcontextprotocol/servers/tree/main/src/slack

Don't install all 6 at once. Install Drive + GitHub on day 1. Add the rest as you discover you need them. Full MCP directory: github.com/modelcontextprotocol/servers

Step 6: Schedule Tasks / Run Managed Agents

Claude Code has a /loop command that lets you schedule recurring tasks. Once set, Claude runs them in the background even when you're not at the computer. This is how you go from "using Claude" to "Claude works for you."

The /loop syntax:

/loop every 30m "check email, triage, draft replies"
/loop @7:30am "generate today's content brief"
/loop weekdays @9:00am "review yesterday's outbound deal activity"

To list or cancel scheduled tasks, ask Claude in plain English ("show my scheduled loops", "cancel the email triage loop") or see the /loop reference in the CLI docs: docs.claude.com/en/docs/claude-code/cli-reference

4 agents worth setting up day 1:

• Email triage (every 30m): scan inbox, label by priority, draft replies for VIPs
• Morning brief (7:00am daily): pull calendar + inbox + tasks, generate one Slack-ready block
• Content review (weekly, Friday 4pm): scan your top-performing posts, draft next week's content angles
• Expense sort (daily): pull new transactions from your bank/CC export, categorize, flag weird ones

Set it once, Claude runs while you sleep. One morning brief automation saves roughly 30 minutes a day. One email triage saves roughly 1 hour a day. Set both up the first weekend and you've already paid back the time it took to install Claude Code.

Step 7: Talk To It Like An Employee

The single biggest mistake non-coders make: they prompt Claude like they're writing a coding spec. Function names, exact specs, line numbers, library versions. Claude already knows all that. What Claude does NOT know is what outcome you actually want.

Don't say (too technical) vs do say (plain English):

• "create a Python def with 3 params" becomes "build me a meal planner"
• "use Anthropic SDK v0.4.2" becomes "email me groceries every sunday"
• "implement it as a singleton" becomes "tell me when it works"
• "return type Promise<User[]>" becomes "give me back a list of users"
• "wrap it in try/catch with retry logic" becomes "if it fails, try again and tell me"

3 prompt patterns to memorize:

OUTCOME + CONSTRAINT: "Build X. Constraint: must work without me logging in every day."

OUTCOME + PROOF: "Build X. Tell me how I'll know it's working when you're done."

OUTCOME + DECISION: "Build X. Before you write code, show me the 3 ways you could approach this and recommend one."

Treat Claude like a senior engineer on your team. You wouldn't tell a senior engineer which Python version to use. You'd tell them what to build, why it matters, and how to know they're done.

Step 8: Ship Your First App

By Monday morning, you'll have a deployed app. Real URL. Built with Claude Code in about 12 minutes once your CLAUDE.md and MCPs are set up. The pattern works for any small personal tool: meal planner, expense tracker, daily journal, weekly OKR review, custom RSS reader.

1. Objective. Open Claude Code and tell it what you want. One sentence: "Build me a meal planner that emails me the grocery list every Sunday at 9am."
2. Structure. Claude proposes the plan: tech stack (Next.js + Vercel + your Gmail MCP), file structure, key files. You approve or request changes. 2 minutes.
3. Execute. Claude writes the code, tests it locally, and shows you the live preview. If something breaks, you say "fix it." No code spelunking required.
4. Ship. Claude deploys to Vercel with one command, sets up the cron job for the weekly email, and gives you a live URL. Total time from idea to deployed: about 12 minutes.

Example session transcript:

you> build me a meal planner that emails groceries every sunday at 9am.
     Use my preferred meals from CLAUDE.md.
claude> Plan:
  - Next.js app deployed to Vercel
  - Sunday cron job hits Gmail MCP
  - Stores meal preferences in a JSON file
  - You confirm meals weekly via a single tap link
you> looks good. ship it.
claude> Building...
  scaffolded Next.js project
  wired Gmail MCP
  deployed to vercel.com/your-name/meal-planner
  cron job scheduled: every Sunday 9:00am
  Live: https://meal-planner.vercel.app
  Test email sent. Check your inbox.

The first time you ship a working app in 12 minutes, something clicks. You stop thinking "I should learn to code" and start thinking "what should I build next." That's the shift this guide is for: from "I use AI" to "I build with AI."

Free hosting: Vercel's free tier handles every personal app you'll build (vercel.com/signup). Starter templates: github.com/coopersimson96/personal-software-examples (5 working personal apps to clone, modify, deploy).

Go Deeper

Printable PDF version of this guide: https://drive.google.com/file/d/1XL4_AZkq5Hv782RQcbHIU8MdRYs9vuAf/view?usp=drivesdk

Want the whole system installed live, with Cooper on the call? Build Your Own AI, the two week sprint runs the first two weeks of August.

Or get a new working tool every Monday inside The AI Operator community on Whop.