Connecting AI to the Business You Already Have

In my last article, “I Had a Business Question. We Didn’t Have a Report for It.,” I described using Claude to explore a sales question. Creating it took about ten minutes. What made that possible was a connection we’d built between Claude and Hyve, our internal business system at Beezwax.

That connection deserves an article of its own. Plenty of businesses have useful information in their existing systems, but getting at it still means finding the right report, exporting something to Excel, or asking someone who knows where to look. Being able to ask a question directly could save a lot of effort. First, though, the AI needs a way to reach that information and enough context to use it properly.

For Hyve, we built a custom MCP connection. It’s also work we’re helping clients with, because the systems they already have are often a good place to start making AI useful in their business.

What is MCP?

MCP stands for Model Context Protocol. It provides a standard way for AI applications to connect to external information and tools. In our case, it’s how Claude accesses information in Hyve. About MCP

From my perspective as someone using it, that means I can ask Claude to help me explore a business question using information from our own system. I still need to think about what I’m asking and check the results, but I have another way to work with that information.

The custom part is important. MCP provides the way to communicate, but someone still has to build the connection to the business system and decide what it makes available. There’s work involved in understanding the system, the information it holds, and what people should be able to do with it.

Your system already knows quite a lot about your business

If you’ve spent years developing a business system, there’s probably a great deal of knowledge built into it. How a customer relates to a project. What counts as an active opportunity. How work moves between departments. Which exceptions need special treatment because, for your business, they happen all the time.

Some of that knowledge is in the data, some is in the application’s logic, and some may still be in the heads of the people who use it. Making it available to AI requires understanding where it lives and how it fits together.

That’s why I think existing systems deserve more attention in these conversations. A business may already have much of the information it needs to do something useful with AI. The next piece of work could be making that information accessible, rather than moving everything into a new application.

That was my experience with Hyve. We’d already done the work of building a system around how Beezwax operates. Connecting Claude gave me another way to use it, including creating a dashboard nobody had needed to specify in advance.

A connection needs some understanding behind it

Suppose you ask, “How are sales doing?” You might mean signed contracts, invoices, or money received. Each could give you a different answer, and each could be correct on its own terms.

Someone who works with you may know what you usually mean. An AI application needs that context to be available somehow. If it makes the wrong assumption, you can end up discussing a convincing answer that has very little to do with the question you thought you asked.

Building a useful connection involves working through those details. What does this information represent? Which source should be used? Are there records that should be excluded? Does the answer need an explanation of how it was calculated?

Clear definitions help us check the result; they don’t guarantee that every AI answer will be right. This is familiar territory for anyone who has built business software. We’ve always needed to understand what the numbers mean before putting them in a report. Giving people more freedom to ask their own questions makes that understanding even more important.

Who gets to ask what?

Access is another part of the work. Someone may be allowed to see the status of a project without being allowed to see its financial details. Connecting an AI application shouldn’t give them a way around that distinction.

We need to decide what the connection makes available and make sure the appropriate permissions are enforced. There’s also a difference between retrieving information and changing it. Looking up a record and updating it have different consequences, and any ability to make changes needs deliberate consideration.

These are practical decisions about how the business wants to use the connection. Starting with a limited set of useful questions can make them easier to work through. You can check the answers against what you already know, see where context is missing, and understand what people actually find useful before expanding it.

Start with something you want to know

I wouldn’t start by asking how to connect everything in the company to AI. I’d start with a question that takes too much effort to answer today.

Perhaps someone puts together the same information from several places every week. Perhaps there’s a report that gets you most of the way there, but you always end up exporting it and doing more work. Or perhaps, as in my case, you want to explore something once and can’t justify asking someone to build it.

A specific question gives us something to work with. We can identify the information needed, look at how it’s currently accessed, and work out what a connection would need to provide. We can also check whether the result is useful, rather than judging success simply by whether the AI can connect.

From there, people can try related questions or create tools around what they need. Some will answer a question once. Others may become part of how a team works. If someone has already built something useful, we can start there and work out what information or connections would help them take it further. If nobody has the time to build it themselves, we can work through the question with them.

Where we come in

At Beezwax, we’re building custom MCP connections as part of helping clients make better use of their existing systems. That means understanding the business as well as building the connection. We need to know what the information means, how the systems fit together, and what access is appropriate for the people using them.

We’re excited to be working this way because clients can do more for themselves once those pieces are in place. They can explore a question, try an idea, and show us what they want to take further. We have something concrete to discuss, and they’ve already had a chance to find out whether it helps.

The dashboard was one small use of the connection we built for Hyve. I deleted the dashboard, but the connection is still there for the next question. That’s the part worth investing in: giving people a useful way to work with the business information they already have, whether they need an answer for half an hour or a tool the whole company will use.