How to Build an MCP Server (Plain English Guide)

There's a technology sitting at the core of everything I've built with AI over the past two years. It's not a product you can buy. It's not an app. Most business owners have never heard of it.

It's called MCP — Model Context Protocol. And once you understand it, everything about how AI connects to your business clicks into place.

This is a plain English guide. Not a code tutorial. If you're a business owner trying to understand what MCP is, why it matters, and how it could work for you — this is for you.

---

What MCP Actually Is

MCP stands for Model Context Protocol. It was created by Anthropic (the company behind Claude). In simple terms, it's a standard way for AI to connect to external tools and data.

Think of it like USB for AI.

Before USB, every device had its own proprietary connector. Printers had one cable. Cameras had another. Keyboards had another. USB standardised the connection. One port, any device.

MCP does the same thing for AI. Before MCP, if you wanted AI to read your email, you needed a custom integration. If you wanted it to check your calendar, another custom integration. Each connection was bespoke, fragile, and expensive to build.

MCP standardises the connection. One protocol, any tool. Your AI can talk to your email, your calendar, your database, your e-commerce platform, your CRM — all through MCP.

---

Why Business Owners Should Care

Here's the practical reality: AI in a browser can answer questions. AI connected to your business systems via MCP can do work.

The difference between "What's a good subject line for my newsletter?" and "Draft next month's newsletter from my running notes, pull subscriber stats from Klaviyo, and schedule it for Thursday" — that difference is MCP.

Without MCP, AI is a clever advisor sitting in a room with no phone and no computer. With MCP, AI has access to every tool you use.

I run six brands. Every one of them is connected through MCP servers. My AI can:

• Read and draft emails (Gmail)
• Check and create calendar events (Google Calendar)
• Pull sales data and manage products (Shopify)
• Send newsletters and track metrics (Klaviyo)
• Create and update tasks (Airtable)
• Search and store files (Google Drive)
• Query a knowledge base of everything I've learned (Open Brain)
• Post to social media (Buffer, Bluesky, X)
• Generate and review content (in my specific brand voice)
• Manage projects across all six brands simultaneously

That's not six separate AI tools. It's one AI — Claude Code — connected to everything through MCP.

---

My MCP Journey

I started with one tool. A single MCP connection that let Claude read and write to a local knowledge base. That was it. One tool, one connection.

Two years later, I run 50+ MCP tools across six brands. But the growth was organic. Each tool was built to solve a specific problem:

• I was spending an hour every morning triaging email → built Gmail MCP tools
• Board reports took two days to compile → built Shopify + Klaviyo + Airtable MCP tools that auto-generate them
• Voice notes were dying in my phone → built a transcription pipeline with MCP tools that create tasks and brain entries
• Customer research was scattered across documents → built a brain query tool that searches everything

Every MCP tool I built replaced a manual process. None of them were built because the technology was cool. All of them were built because a specific pain point demanded a solution.

---

How MCP Works (Conceptual Guide)

You don't need to understand the code. You need to understand the architecture. There are three things an MCP server can provide:

Tools

Tools are actions the AI can take. "Send an email." "Create a task." "Pull last month's sales." Each tool has a name, a description, and a set of inputs. When the AI decides it needs to send an email, it calls the email tool with the recipient, subject, and body.

Think of tools as buttons on a control panel. Each button does one specific thing. The AI reads the labels and presses the right button for the job.

Resources

Resources are data the AI can read. Your brand voice document. Your product catalogue. Your project status. The AI doesn't modify resources — it reads them for context.

Think of resources as reference books on a shelf. The AI pulls down the right book when it needs background information.

Prompts

Prompts are pre-written instructions for specific tasks. "Generate a board report" might be a prompt that tells the AI exactly which data to pull, what format to use, and what voice to write in.

Think of prompts as standard operating procedures. Instead of explaining the process every time, the AI follows the documented procedure.

---

What You Can Connect

Essentially anything with an API. Here's what I've connected in practice:

Email: Gmail for two separate Google Workspace accounts (personal farm business + consulting practice). Read inbox, draft replies, search, archive, triage, create labels.

Calendar: Google Calendar for both accounts. View today's schedule, create events, check upcoming week.

E-commerce: Shopify store — products, orders, inventory, customers, discounts, analytics.

Email Marketing: Klaviyo — campaigns, templates, subscriber lists, segments, metrics. Draft, create, and send newsletters.

Project Management: Airtable — tasks, projects, strategy logs, weekly reviews. Full CRUD (create, read, update, delete) across all tables.

File Storage: Google Drive — search files, read documents, create files, mirror content. Separate Drive access for farm and consulting.

Knowledge Base: Open Brain — a living knowledge base that stores everything I've learned across all projects. Query it by topic, ingest new information, track patterns.

Social Media: Buffer for scheduling, Bluesky for direct posting, X (Twitter) for threads and posts.

Automation: n8n workflows for complex multi-step automations. Trigger webhooks, check execution status, activate/deactivate workflows.

Conservation: TrapNZ for predator trapping data. Lines, traps, catches, project stats.

Payments: Stripe for product creation and payment links.

Communication: Telegram bots for mobile access to the entire system.

---

The Architecture: STDIO vs HTTP

There are two ways to run an MCP server. You don't need to build either yourself, but understanding the difference helps you make decisions:

STDIO (local): The MCP server runs on your computer. Claude Code communicates with it directly. Fast, private, no internet required for the connection itself. This is how most of my tools run — on my Mac Mini, talking directly to Claude Code.

HTTP (remote): The MCP server runs on a server somewhere (Railway, Fly.io, a VPS). Claude Code connects to it over the internet. Useful when you need the MCP server accessible from multiple devices or when it needs to run 24/7 independently.

I use STDIO for most things. HTTP for the few tools that need to be always-available or shared across machines.

---

5 MCP Tools I Use Most

Out of 50+ tools, these five get used daily:

1. Brain Query / Brain Ingest

The Open Brain is my living knowledge base. query_brain searches it by topic — "what do we know about Shopify shipping rates?" — and returns relevant entries with context. ingest_to_brain saves new knowledge — decisions, meeting notes, project updates, strategy insights.

This is the most valuable tool I've built. It means nothing gets lost. Every decision, every insight, every piece of research is searchable and persistent across sessions.

2. Gmail Inbox / Triage

gmail_inbox shows me unread mail. gmail_triage categorises everything into Action Required, Leads, FYI, and Noise. I say "check email" and get a sorted inbox. I say "draft a reply" and Claude writes one in my voice. I review and send.

3. Calendar Today / Upcoming

"What's on today?" pulls my full schedule. "What's coming up this week?" gives me the next seven days. Simple, but it means I never context-switch to a calendar app during a work session.

4. Task Management

create_task and list_tasks connect to Airtable. Every task has a project, status, priority, due date, and client tag. I can say "create a task: fix the checkout flow on the Shopify store, link it to Shop & Product Management, high priority" and it's done. The task shows up on my live dashboard at buildwithbilly.ai.

5. Content Generation

generate_copy produces social posts, newsletter drafts, and blog outlines in the specific brand voice of whichever project I'm working on. It pulls the brand voice document, reads recent content for tone calibration, and drafts. I edit and publish.

---

When to Build Your Own vs Use Existing MCPs

There are now hundreds of pre-built MCP servers available. Anthropic maintains official ones for common services. The community has built many more.

Use existing MCPs when:

• The service is common (Gmail, Slack, GitHub, databases)
• You need standard functionality (read/write/list)
• You want to get started quickly

Build your own when:

• You need custom business logic (our board report tool combines data from 5 sources with specific formatting)
• You're connecting to a niche service (TrapNZ doesn't have a public MCP)
• You want tight integration with your specific workflows
• You need to enforce business rules (our content tools enforce brand voice, our task tools enforce project linking)

Most businesses should start with existing MCPs and only build custom ones when they hit limitations. The exception is the knowledge brain — everyone's brain is unique, and building your own is where the real leverage lives.

---

The Open Brain: An MCP-First Product

Everything I've learned about building MCP-powered knowledge systems is packaged into the Open Brain.

It's a starter kit that gives you:

• A working knowledge brain you can query and grow
• The MCP server architecture to connect it to Claude Code
• Templates for brain pages (project notes, meeting notes, strategy logs)
• The ingestion pipeline that turns voice notes, documents, and conversations into searchable knowledge

This isn't a SaaS product that charges you monthly. It's a system you own and run on your own machine. No lock-in, no data leaving your control, no subscription.

The Brain Starter Kit is $27 and gets you started in an afternoon.

Get the Brain Starter Kit

---

The Business Case for MCP

Here's the honest ROI calculation:

Time saved: I estimate MCP tools save me 15-20 hours per week across all brands. That's email triage, report generation, task management, content drafting, and knowledge management that used to be manual.

Cost: The MCP servers themselves cost nothing to run (STDIO on my local machine). The services they connect to have their own costs (Shopify, Klaviyo, etc.) but those existed before MCP. The marginal cost of MCP is essentially zero.

Quality: My newsletters are more consistent because the AI references our voice document every time. My board reports are more comprehensive because the AI pulls from every data source. My task management is cleaner because everything goes through a structured pipeline instead of sticky notes and forgotten Slack messages.

Compounding knowledge: Every piece of information that passes through MCP gets captured in the brain. Six months of accumulated knowledge means the AI's responses get better over time. It remembers decisions, learns context, and builds on previous work.

---

Getting Started

If you're a business owner who wants to connect AI to your actual business tools:

• Start with Claude Code. It's the foundation. Read my guide on Claude Code for business.

• Pick one pain point. Don't try to connect everything at once. What's the one thing that wastes most of your time?

• Check for existing MCPs. Chances are someone has already built a connector for the tool you need.

• Consider a brain. A knowledge base that grows with your business is the highest-leverage MCP tool you can build. The Brain Starter Kit is a good starting point.

• Or skip the DIY and talk to someone. I offer consulting specifically for business owners who want MCP-powered AI systems without the learning curve.

Want to see what Claude Code skills look like in practice? Check out the Claude Code Skills Pack.

---

Want to know where AI fits before you start building?

Take the free AI Time-Savings Calculator — it takes 2 minutes and shows you the hours you're losing to manual work each week. That's the starting point for knowing which MCP tools to build first.

Stay human, Billy.