The Claude Code Unlock Checklist for Business Owners

How to Use This Checklist

Most people install Claude Code, ask a few questions, and move on. A post from @mertmetindev on X put it plainly: most people install it, type 2 commands, and use 10% of its power. That post got 491 likes. People recognized themselves.

I was one of them. I spent two weeks on basic prompts before I looked at the configuration. What I found were 4 layers that most business owners never set up. Once I had them running, my work changed.

This checklist covers those 4 layers. Work through them in order. Each one builds on the last.

• Check off each item as you complete it.
• You don't need to finish everything in one session.
• Layer 1 alone will improve every Claude Code session you run from that point forward. None of this requires coding experience.

---

Layer 1: Your CLAUDE.md File (~30 minutes)

CLAUDE.md is a plain text file that lives in your project folder. Claude reads it at the start of every session. It tells Claude who you are, how you work, and what to avoid.

Without it, Claude makes assumptions and defaults to generic output. With it, every session starts with your context already loaded.

Checklist:

• ☐ Create a file named CLAUDE.md in your main project folder
• ☐ Write your business name and what you do in 2-3 plain sentences
• ☐ Describe who your customers are, as a real person with a real situation, not a demographic category
• ☐ List your voice rules: words you use, words you never use, how formal or casual you write
• ☐ Note what you're working on right now (update this monthly)
• ☐ Set output preferences: short or detailed, bullets or paragraphs, what you never want to see
• ☐ List anything Claude should never do in your specific work context

---

Layer 2: Your First Hooks (~1-2 hours)

Hooks are small scripts that run automatically. You don't type a command. They fire at the start of a session, the end of a session, or after a set number of turns.

Here are the 4 hooks I run:

• A turn counter that fires after every 15 turns and reminds me to save important decisions
• A Stop hook that checks git push is allowed before it tries to deploy
• A PreToolUse hook that logs the date and project name at the start of each session
• A memory reminder that fires when sessions run long

None of these happen when I type a prompt. They just run.

A post by @RoundtableSpace on X (274 likes) showed a plugin that auto-installs hooks and skills in one step. People were excited because they didn't know hooks existed before they saw it. That's how many people are running Claude Code without them.

Checklist:

• ☐ Read through Claude Code's hook documentation once to understand the timing options (Stop, PreToolUse, PostToolUse, UserPromptSubmit)
• ☐ Write one hook: a reminder to save important decisions after every N turns
• ☐ Test it in a live session to confirm it fires correctly
• ☐ Add a second hook only after the first one is working

---

Layer 3: Your First Skill (~2-3 hours)

Skills are saved instructions for a task you do more than once. You give the skill a name, write the steps, and trigger it with one command.

Skills I run:

• /last30days: research across Reddit, X, and YouTube simultaneously
• /linkedin-post-automation: 7 posts, 7 images, Buffer scheduling in one run
• /deploy: commit, push, and put the website live

Each of those was once a series of steps I typed manually. Now one command runs the whole sequence.

Checklist:

• ☐ List 3 tasks in your work that you repeat regularly
• ☐ Pick the one with the clearest step-by-step pattern
• ☐ Write the skill: describe each step in plain English, including what files or tools it uses
• ☐ Test it twice. Fix what breaks.
• ☐ Add your second skill only after the first is working reliably

---

Layer 4: Your First MCP Connection (~2-4 hours)

MCP servers are connections between Claude and your real business tools. Instead of copying data out of one system and pasting it into a prompt, Claude reads and writes directly to the tool.

My setup connects Claude to:

• Buffer's GraphQL API to schedule LinkedIn posts automatically
• Cloudinary to store and serve images
• GitHub to commit and push website changes
• SiteGround for deployment

Before these connections, every workflow ended with manual handoffs. Copy the post text, paste it into Buffer. Download the image, upload it again. Now Claude calls each tool directly.

Any tool with an API can have an MCP server built for it. For common tools like HubSpot, QuickBooks, Airtable, and Google Sheets, servers already exist. For anything custom, a developer can usually build one in a few hours.

Checklist:

• ☐ List the tools you currently copy data out of manually at the end of a workflow
• ☐ Pick the one tool you connect to most often
• ☐ Search for an existing MCP server for that tool (Claude Code's documentation has a server registry)
• ☐ If one exists, set it up and test with a read-only operation first
• ☐ If none exists, talk to a developer about building one. Bring the API documentation for your tool.

---

Where to Start

Set aside time for one layer per week. That's the pace that works.

You don't need all four before you get value. Layer 1 alone changes the output quality from the first session.

---

---