Automate Your Business with Claude:
The Complete 5-Level Setup Guide

💡 How to use this guide: Read Section 0 first. Find your level. Jump directly to that section. You don’t need to read every level - start where you are and move up when you’re ready.

Before You Start: Prerequisites

⚠️ Note: Levels 1 and 2 work on the free tier. Levels 3 and 4 require a paid plan (Pro or higher). Level 5 (API) requires a separate API key from console.anthropic.com - this is billed per token, separate from your subscription.

Section 0 - Level Selector: Which Level Are You?

The 5-Level Quick Reference

2-Question Decision Flowchart

Do you want Claude to run automatically WITHOUT you opening it or prompting it?
│
├── NO → Do you want Claude to remember your business context automatically?
│         ├── NO  → Start at Level 1 (Connect)
│         └── YES → Start at Level 2 (Projects)
│
└── YES → Are you comfortable using a terminal or a no-code tool like Zapier / Make?
          ├── NO (but comfortable with desktop apps) → Start at Level 3 (Cowork)
          ├── YES (terminal / code)                  → Start at Level 4 (Claude Code)
          └── YES (Zapier / Make / n8n)              → Start at Level 5 (API)

Section 1 - Level 1: Connect Your Tools & Delegate

What This Actually Means

At Level 1, Claude acts as a central hub between the apps you already use. You connect your tools (Gmail, Notion, Slack, Google Drive, Google Calendar, HubSpot) via MCP connectors - and then ask Claude to do things across them in a single prompt.

You still initiate every action. Claude doesn’t run on its own. Think of this as replacing tab-switching and copy-pasting with one conversation.

What Level 1 Is NOT

• It is not autonomous. Claude waits for you to ask.
• It does not run in the background.
• It does not remember your business context between chats (that’s Level 2).

How to Connect Your Apps (Step by Step)

1. Go to claude.ai → click your profile icon → Connectors
2. Browse the connector directory (300+ available as of 2026)
3. Click Connect next to the app you want
4. Authorise Claude’s access via OAuth (same as any standard app login)
5. Done - Claude can now read and write to that app in any chat

App Connection Table

4 Ready-to-Use Prompts (Copy-Paste)

Section 2 - Level 2: Projects = Your Reusable Brain

The Problem This Solves

Every time you open a new Claude chat, it knows nothing about you. You paste your brand voice. You re-explain your business. You describe your tone. You do it again the next day. And the next.

Claude Projects solve this. You load your context once. Every conversation inside that project inherits it automatically. Claude already knows who you are, who you serve, and how you communicate - before you type a single word.

What a Project Actually Contains

💡 RAG Mode: On paid plans, Projects automatically activate Retrieval Augmented Generation (RAG) when your uploaded content gets large. This expands your effective knowledge base by up to 10x while maintaining response quality. You don’t set this up manually - it happens automatically.

How to Set Up a Project (Step by Step)

1. Go to claude.ai → click Projects in the left sidebar
2. Click New Project → give it a clear, functional name (e.g. “Content Engine” or “Client: [Name]”)
3. Click Add Project Instructions → write what Claude should know and how it should behave
4. Click Add to Project Knowledge → upload your documents (PDF, DOCX, TXT, CSV, HTML - up to 30MB per file)
5. Start a new chat inside the project - Claude now has everything it needs

What to Upload: The Context Loading Table

• Brand voice guide (PDF/TXT): Every piece of output sounds like you without reminders
• SOPs (PDF/DOCX): Claude follows your exact processes for repeatable tasks
• Offer/service descriptions (TXT/DOCX): Claude pitches and explains your services accurately
• Best-performing past content (TXT): Claude mirrors what’s already worked in your voice
• Target audience personas (PDF/TXT): Claude writes for the right person every time
• Banned phrases / style rules (TXT): Claude never uses the words or phrases you hate
• Competitor research (PDF): Claude references market positioning in content and strategy
• FAQ document (DOCX/TXT): Claude answers client questions without you writing them

💡 Pro tip: Name your files descriptively. When prompting Claude, reference them explicitly: “Using brand-voice.md and offer-descriptions.txt, write a LinkedIn post about [topic].” Output quality improves immediately.

Project Instruction Template (Copy-Paste and Customise)

You are [role] for [business/brand name].

BUSINESS CONTEXT:
- What we do: [one sentence]
- Who we serve: [ICP description]
- Our core offer: [offer description]

VOICE & TONE:
- Always write in [tone: direct / warm / professional / conversational]
- Sentence style: [short and punchy / flowing / varies by context]
- Words we never use: [list]
- Words we love: [list]

BEHAVIOUR RULES:
- Always [specific rule 1]
- Never [specific rule 2]
- When asked for [content type], always [specific format rule]

FILES IN THIS PROJECT:
- brand-voice.md: Use this for tone guidance on every output
- sops.pdf: Reference this for any process questions
- offers.txt: Use this when describing our services or writing pitches

Section 3 - Level 3: Cowork = Hand Off the Whole Task

What Cowork Actually Is

Claude Cowork is a desktop tool (available via the Claude Desktop app) that gives Claude access to files and folders on your computer. You describe a multi-step job. Claude works on it - reading files, running code, organising content, drafting deliverables - while you step away.

Cowork reached general availability in April 2026 across all paid plans (Pro, Max, Team, Enterprise) on macOS and Windows.

⚠️ Important: Your desktop app must stay open and your computer must stay awake while Claude works. If you close the app, active tasks stop.

How to Get Started

1. Download Claude Desktop from claude.ai/download (free)
2. Sign in with your Claude account (Pro or higher required for Cowork)
3. Open the Cowork tab in the sidebar
4. Select a folder on your computer to give Claude access to
5. Type your task - Claude analyses your request, creates a plan, and begins working

What Cowork Handles Best

How to Write a Cowork Task Brief (The Framework)

A good Cowork brief has 4 components:

• GOAL: [What is the final deliverable?]
• INPUT: [What files or information does Claude have access to?]
• PROCESS: [Any specific steps or approach to follow?]
• OUTPUT FORMAT: [How should the result be structured? File type? Length?]

Section 4 - Level 4: Claude Code Builds the Machine

What Claude Code Is

Claude Code is Anthropic’s agentic command-line tool. It lives in your terminal. You describe what you want in natural language. Claude reads your files, writes and executes code, connects to external tools, and builds automation pipelines - all from natural language instructions.

💡 You don’t need to be a developer. Claude Code handles the code itself. You describe what you want, Claude builds it. The term for this is “vibe coding” - non-programmers are now building production-grade automations using Claude Code every day.

Installing Claude Code

macOS / Linux / WSL

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

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex

Or via npm (requires Node.js 18+)

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

⚠️ Do NOT use sudo npm install -g — this causes permission issues. If you hit permission errors, see the troubleshooting guide at docs.anthropic.com.

After installation, verify with:

claude doctor

To start Claude Code in any folder:

cd your-project-folder
claude

The 4-Stage Automation Pipeline

Every Claude Code automation follows the same structure:

TRIGGER → COLLECT → TRANSFORM → OUTPUT

Connecting Claude Code to Your Other Tools (MCP)

Claude Code connects to external tools via MCP (Model Context Protocol). This lets Claude call tools like Clay, HubSpot, Notion, Slack, and more — directly from your automation scripts. See Anthropic MCP docs for the full connector list.

To add an MCP connection, create or edit the .claude/mcp.json file in your project:

{
  "mcpServers": {
    "notion": {
      "url": "https://mcp.notion.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_NOTION_API_KEY"
      }
    },
    "slack": {
      "url": "https://mcp.slack.com/mcp"
    }
  }
}

Then, inside Claude Code, you can say: "Read my open Notion tasks and post a summary to the #team Slack channel."

Building a Reusable Skill (SOP for Your Agent)

A Skill is a saved instruction file that tells Claude Code exactly how to run a repeatable process. Think of it as an SOP for your AI agent. Create a file called skill.md in your project directory:

---
name: Weekly Content Report
description: Reads the week's content metrics and drafts a summary report every Monday
---

## Steps
1. Read all files in the /metrics/weekly folder
2. Extract: total reach, top-performing post, engagement rate, follower change
3. Compare to the previous week's file in /metrics/weekly
4. Write a 200-word narrative summary of performance
5. Flag any metric that dropped more than 15% week-over-week
6. Save the report as YYYY-MM-DD-weekly-report.md in /reports

Section 5 - Level 5: API = Set It and Forget It

What the API Actually Means (Plain Language)

The Claude API is Claude running silently inside another tool. You are not opening Claude. You are not typing prompts. Something else triggers Claude, Claude does its job, and the result goes directly to where it needs to go - your CRM, your email inbox, your Slack, your spreadsheet.

External trigger → Your automation tool (Zapier/Make/n8n) → Claude API (processes it) → Output destination (HubSpot/Slack)

Which Tool to Use: Zapier vs Make vs n8n

Recommendation: If you’ve never automated before, start with Zapier. It has native Claude integration. If you want more control or lower costs at scale, use n8n.

Setting Up Claude in Zapier (Step by Step)

1. Go to zapier.com → click Create Zap
2. Set your trigger (e.g. HubSpot: New Contact Created)
3. Add an action step → search for Anthropic (Claude)
4. Select Send Message → connect your Anthropic account by pasting your API key
5. Choose your model (Haiku 4.5 for high volume, Sonnet 4.6 for most tasks)
6. Write your system prompt (this is where output quality is determined)
7. In the user message field, map in your trigger data (lead's name, company, form response)
8. Add your output step (e.g. send to Gmail, update HubSpot, post to Slack)
9. Test and publish

Setting Up Claude in n8n (HTTP Request Node)

Method: POST
URL: https://api.anthropic.com/v1/messages

Headers:
  x-api-key: YOUR_ANTHROPIC_API_KEY
  anthropic-version: 2023-06-01
  Content-Type: application/json

Body (JSON):
{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1000,
  "messages": [
    {
      "role": "user",
      "content": "{{ $json.your_trigger_data }}"
    }
  ]
}

Extract response with: {{ $json.content[0].text }}

A Complete Working Example: New Lead → Qualify → Route

Scenario: Someone fills out your website contact form. You want Claude to qualify them, score them, and if they're a fit, send a personalised first-touch email within 5 minutes - automatically.

Zap structure:
TRIGGER: Typeform - New submission
          ↓
ACTION 1: Anthropic (Claude) - Qualify the lead
  System prompt: "You are a lead qualifier. Score this lead 1-10 on ICP fit
                  based on the form responses below. Return JSON:
                  {score, tier, reasoning, personalised_opener}"
  User message: Name: {{name}}, Company: {{company}},
                Role: {{role}}, Use case: {{form_answer}}
          ↓
ACTION 2: Filter - Only continue if score >= 7
          ↓
ACTION 3: Gmail - Send personalised email
  To: {{email}}
  Body: Uses {{personalised_opener}} from Claude's output
          ↓
ACTION 4: HubSpot - Create contact + set Lead Score field to {{score}}
          ↓
ACTION 5: Slack - Post to #hot-leads with summary

Section 6 - The "What to Automate First" Framework

The Simple Rule: Start where the rules are clear and the volume is high.

If a task happens frequently, follows a predictable pattern, and produces a specific output - automate it now. If a task requires judgment, nuance, or relationships - keep yourself in the loop.

The Task Audit: Find Your First Automations

Score each of your top tasks on 3 dimensions (1 = low, 3 = high). Tasks with the highest total score are your best automation candidates. Score → Recommendation: 7-9: Automate now, 4-6: Automate with human review, 1-3: Keep human.

The 5 Tasks Most Business Owners Should Automate First

Section 7 - Quick Reference: Prompt Templates by Level

LEVEL 1 - Connected Tools

LEVEL 2 - Projects

LEVEL 3 - Cowork

LEVEL 4 - Claude Code

LEVEL 5 - API (System Prompts)

Stack Recommendations by Business Type

Frequently Asked Questions