Your data in one place, reading and writing business intelligence. If you’re using Claude or ChatGPT and finding it useful but limited, here’s the missing piece: AI gets dramatically more powerful the moment you connect it to your own data.

If you connect Claude to your business data with MCP directly, you can ask it questions and get answers on what’s actually happening right now in your business. The key is a workflow that SA AI specialist Theunis Duminy teaches founders in his masterclass on building a Founder OS.

Theunis is the founder of Vertiqal, an SA AI implementation consultancy and training studio. He spends most of his time helping companies of all sizes (from solo founders to large enterprises) connect AI to the actual systems where their data lives. The pattern he sees over and over: Founders treating AI as a search box when it could be running an interface to their entire business.

Here’s his workflow for connecting Claude to your business data using MCP.

MCP (Model Context Protocol) is a standard that lets AI tools like Claude connect directly to external systems. Databases, file storage, communication tools and project trackers. The AI gets read and write access (with permissions you control), and suddenly the conversation changes.

Instead of “here’s my pipeline, what should I do?” it becomes “what’s in my pipeline this week?” and Claude actually goes and looks.

As Theunis puts it, “Your interface for talking to your data really becomes Claude. Once you’ve set this up once, you should minimally go back to the actual interface of your database — you can just use Claude to interact with it.”

That’s the unlock. AI stops being a separate tool you visit and becomes the interface to the tools you already have.

Before you connect Claude to anything, your data needs to live somewhere it can actually be queried. If your client list is in your head, your deals are in WhatsApp, and your tasks are in three different apps, MCP won’t fix that. It will just give you a chat interface to chaos.

Theunis is emphatic about this: “Centralise your data first. Everything else (AI, automations, reporting) becomes trivially easy after that.”

In his system, that centralised place is Airtable, because it works like a relational database with a spreadsheet interface. But the principle holds for any database. The tool matters less than the structure: one place where your companies, contacts, deals, projects and tasks all live and link to each other.

If you’re not there yet, fix that before you do anything else with MCP.

MCP only works through Claude Desktop, the standalone app, not the web version. Download it, sign in, then look for the connectors section in your settings.

This is where Claude lists every external system it can connect to. There are dozens of MCP servers available now: Airtable, Google Drive, Slack, GitHub, your file system, calendars and more. Each one is a separate connection with its own permissions.

For your first MCP connection, pick the system where your most important business data lives. If you’re using Airtable as your central database, start there. If your data lives somewhere else, find the matching MCP connector and use that instead.

This is the step most founders rush. When you connect an MCP server, Claude asks what it’s allowed to do. Read records. Create records. Update records. Delete records.

Theunis’s default rule: “Everything is okay unless we are deleting, creating or updating specific things. In those cases, I want Claude to just ask me what’s happening.”

In practice, that means: Read access is fine to grant openly, but anything that changes your data should require Claude to confirm with you before acting. This catches mistakes before they happen and keeps you in control of anything that matters.

Spend the extra minute getting permissions right. The first time Claude asks, “Can I update this record?” and you spot something you didn’t intend, you’ll be glad you set it up this way.

MCP connects Claude to your data. A Claude Project tells it what your data means.

Create a new project in Claude Desktop (call it something like “Founder OS”) and use the project instructions to give Claude context. Specifically:

This step is what separates a generic AI chat from an actual business interface. Skip it and Claude will guess. Do it properly and Claude will respond like someone who actually knows your business.

Once it’s set up, test it with something simple. “What tasks are outstanding that I need to focus on this week?” Claude queries your database, pulls the relevant records, and answers. No copy-paste, no context-loading, no tab-switching.

Once read queries are working, try a write. Theunis’s example: “The content pipeline setup deal has now been won. The client is (whoever). Please update accordingly.”

Claude reads the request, finds the deal, prepares the update, and, because of how you’ve set up permissions, asks you to confirm before changing anything. You say yes. The record updates. You never opened the database.

That’s the moment most founders realise this is genuinely different from anything they’ve used before.

Firstly, SA founders typically run leaner stacks than their global counterparts. Fewer apps, smaller teams, less budget for enterprise tooling. That actually makes MCP-style integration more valuable, not less. When you can’t afford five SaaS subscriptions, the ability to connect AI directly to one well-structured database becomes a meaningful operational lever.

Secondly, AI literacy in SA is uneven. Most founders are still using AI as a smarter search box. Setting up MCP puts you ahead of the curve in a way that compounds. The founder who has Claude wired into their business data is operating at a different level of speed than the founder still copy-pasting context into ChatGPT — and that gap will widen as MCP-compatible tools multiply over the next year.

Theunis frames it like this: “One person with the right system can do what previously took possibly 10 people. The limiting factor isn’t budget or team size. It’s whether you’re thinking clearly about where your data lives and how it connects.”

Once Claude is connected to your business data, the way you work changes: You stop opening five tabs to find one piece of information and start asking questions in plain English and getting real answers from your real data. You start updating records by describing what happened instead of clicking through forms.

It takes an afternoon to set up. It saves you the hours every week you currently spend being your own integration layer.

Connecting Claude via MCP is one piece of a much bigger system. In Theunis’s full Founder Collab masterclass on Automations Every Founder Needs in Their Stack: A Practical Founder OS, he walks through:

You can get access to Theunis’s full playbook, plus 40+ other masterclasses from South African operators, founders and experts when you join The Founder Collab.

This workflow first appeared in our 4 May 2026 edition on the Pagamio commerce platform. 

Get more SA tech and business news and subscribe to The Open Letter.