What is MCP? A plain-language guide for business owners.

The problem: your data lives where AI can't reach it

Ask a modern AI assistant a general question and it's remarkably capable. Ask it “which of our invoices are overdue?” and it has no idea — because that answer lives inside your invoicing system, your CRM, your planning tool, maybe a desktop application from 2009. The assistant is smart; it just can't see your business. So people end up copy-pasting between their apps and a chat window, which recreates the manual work AI was supposed to remove.

What MCP is

MCP — the Model Context Protocol — is an open standard for connecting AI assistants to tools and data. It originated at Anthropic in November 2024, and since then it has been adopted broadly across the industry: OpenAI, Google and others announced support during 2025. In practice that means it isn't one vendor's plug format — it's a common language, so a connector you build for your systems works with the major AI assistants rather than locking you to one.

A useful mental model: MCP is a standard socket. Your software gets one fitted, and any compatible AI assistant can plug in — through a door you define, not through the walls.

What a connector does in practice

An MCP connector is a small piece of software that sits next to one of your systems and exposes specific things an AI is allowed to see and do. With one in place, the day-to-day looks like this:

• Someone on your team asks, in plain language: “Which invoices are more than 30 days overdue?” or “Summarise this customer's history before my 2 o'clock call.”
• The assistant reads the answer from your real system through the connector — and, where you've allowed it, acts too: drafting the reminder email, creating the follow-up task, updating the record.
• Everything happens inside limits you set. A connector is permissioned — you decide exactly which data it can read and which actions it can take — and its activity can be logged, so you can see what the AI did and when.

Your existing software stays exactly where it is. Nothing is migrated, nothing is replaced. The connector is the door, and you hold the keys.

What about legacy and desktop software?

This is where it gets interesting for businesses whose critical tools aren't shiny cloud apps. Plenty of companies run on desktop software — often Java applications that have quietly done the job for fifteen years and have no API at all. Conventional wisdom says those can't join the AI era without a rewrite.

They can. As a working, open-source example: we built Swing MCP, a connector that lets AI assistants operate Java Swing desktop applications directly — taking snapshots of the interface, clicking buttons, typing, filling forms, reading tables and trees, working menus and dialogs. It's the kind of “legacy” software most agencies won't touch, connected to AI without changing a line of the original application. The code is public on GitHub, so you don't have to take our word for it.

The broader point: if software runs your business, there is very likely a safe, standard way to let AI work with it — whether it lives in the cloud, on a server, or on the PC in the back office.

What it costs to explore

Nothing, to start. Our free AI Opportunity Audit is a 30-minute call about how your business actually runs and which systems you use; within a week you get a short written automation map — what's worth connecting or automating first, roughly what it takes, and what we'd skip. If a connector is the right move, a first working one typically takes two to four weeks. And if AI isn't worth it for your case, the audit will say exactly that — free advice to not hire us is part of the deal.

← Back to guides