AI Business Intelligence for Counterpoint.
Ask your data anything.

Barton is a secure connector, built on the Model Context Protocol (MCP) — an open standard that lets an AI assistant like Claude securely connect to external systems and databases in real time.

Think of it this way: Claude is extraordinarily intelligent, but out of the box it knows nothing about your business. It can’t see your sales figures, your inventory levels, or your transaction history — because that data lives in NCR Counterpoint, not on the internet.

Barton is the bridge. It gives Claude secure, structured access to your Counterpoint data so when you ask a question, Claude is reading your actual numbers — not guessing. Without it, you’d need a developer to manually export data, write custom queries, or build a full integration from scratch.

But Barton isn’t just a pipe. Counterpoint has decades of database structure behind it — which table holds on-hand quantity, how a transfer differs from a sale, how stock moves between locations, why two fields that look identical aren’t. All of that’s built into Barton, so Claude reads your numbers the way you’d read them instead of guessing at them like a stranger would. That’s the hard part, and it’s the part we’ve spent years learning.

We don’t make AI. We make it worth having if you run Counterpoint.

Claude is a general-purpose AI. It’s brilliant at reasoning and answering questions — but it has no way to access private business systems on its own. It doesn’t know your SKUs, your sales history, or your store’s inventory.

Beyond access, there’s the question of how to speak Counterpoint. NCR Counterpoint has a specific database schema, data relationships, and business logic built up over decades. Knowing which tables to query, how inventory is tracked across locations, and how transactions are structured requires deep domain expertise. That’s the work Barton does, so Claude can interpret your Counterpoint data correctly from the first question you ask.

Yes — Barton is a connector, not a standalone app. You’ll need an active subscription with an AI assistant that supports custom connectors.

We build for Claude, we test against Claude, and Claude is the only assistant we support. Every prompt, every setup instruction and every validation pass we run is against Claude, made by Anthropic — the company that created the MCP standard Barton runs on. It’s the configuration we know performs, and it’s what we recommend without hesitation.

MCP is an open standard, so Barton will technically function with any AI that supports custom MCP connectors, including ChatGPT, Perplexity and others. If that’s already your company standard, it will work — but understand what you’re choosing. We haven’t tested or tuned it there, results will vary, and we can’t troubleshoot it for you. Setup support and accuracy tuning are for Claude configurations only.

Your AI subscription is billed separately, directly by the provider.

Honest answer: very good, but not guaranteed to be perfect. Barton gives Claude direct access to your live Counterpoint data, so the raw numbers it retrieves are accurate. The risk lies in interpretation — how Claude understands your question, which time range it applies, or how it handles unusual data conditions.

Our recommendation: trust, but verify. Use the AI to surface insights quickly, then spot-check anything that will drive a purchasing decision, staffing change, or financial report against your Counterpoint data directly. Think of it as a very fast, very capable first draft — not a final authority.

Not quite, and the reason is worth a minute of your time.

Claude doesn’t have your data sitting in front of it. It works the way a good analyst works: read the question, work out what it needs, go get it, look at what came back — and go back for more if the answer isn’t there yet.

So when you ask “why has this category been slipping?”, Claude doesn’t run one lookup. It asks Barton for the category’s recent sales. Reads them. Realizes it needs last year’s numbers to tell a real decline from a seasonal one, so it asks again. Notices the drop is in one location and not the other three, so it asks again. Every one of those is a round trip to your Counterpoint database.

Simple questions are a single trip and come back in seconds. Analytical ones take several, and run longer, but get you a more comprehensive answer.

It also varies by which assistant you use. Claude leans analytical: it would rather make one more call to Barton than hand you a confident guess. That’s exactly what you want when the number is about to drive a purchase order, and it’s part of why we build for Claude.

Yes, Anthropic allows custom connectors on the free Claude plan, limited to one connector — which is all you need for Barton.

The catch is usage. Free accounts have tight limits on how much you can ask, and you’ll reach them quickly once you start putting real questions to real data. It’s fine for occasional, casual use.

There’s no specific model to choose. Connectors work at the account level, not the model level — so whichever current Claude model you’re working in will be able to use Barton. You don’t need to hunt for the right one.

What matters is the plan. Anthropic supports custom connectors on Free, Pro, Max, Team and Enterprise plans. For most retailers an individual paid plan (Claude Pro, currently $20/month) is the practical starting point. If several people will use it, a Team plan is worth considering — on Team and Enterprise an account Owner adds the connector once and everyone else simply connects to it, which is far easier to administer than setting it up person by person.

Plans and prices are set by Anthropic and do change. Check their current pricing page before you commit, and we’re happy to talk through which plan fits how your team works.

Your data never leaves your environment without you asking for it. Barton operates on a request-only basis — when you ask a question, it fetches only the data needed to answer it. Nothing is stored on our servers, and nothing is sent to Claude proactively.

Claude is governed by Anthropic’s privacy policies, available at anthropic.com. We recommend reviewing both before enterprise deployment.