We Build MCPs and Give Managers and Owners Back Control over AI

Ask any marketing manager what AI is doing in their team right now, and you will get an honest answer: "A lot. But don't ask me exactly what."

Teams paste customer data into chat windows. Every specialist uses their own favourite tool. Recommendations sound convincing, but nobody can trace what information they were based on. The work goes faster, that much is true. But the grip is gone.

That is not an AI strategy. That is a copy-paste economy.

The real problem is not the AI

In an earlier article I explained what MCP is: the Model Context Protocol, the universal plug that lets AI talk directly and securely to your business systems. That article was about the technology. This article is about the question that comes next, and that matters more.

Because the real problem with AI in marketing teams is not that the models aren't smart enough. The problem is twofold:

1. The AI doesn't know your business. Without your context you get industry averages, not advice about your account, your margins, your strategy.
2. You don't know the AI. You don't know which data your team is sharing, which assumptions sit in the output and who is responsible for what.

The first problem makes AI useless. The second makes AI risky. And most organisations suffer from both at the same time.

What a custom MCP changes

At Format FWD we build MCP servers for clients, as part of our MAXai pillar. In plain language: we build the knowledge layer between AI and your reality.

It looks like this:

1. A company brain as the foundation. Identity, brand values, goals, channel strategy and history are captured as permanent context. Every analysis starts from who you are, not from scratch.
2. Connections to the systems that matter. Google Ads, Meta, GA4, your CRM or webshop: one secure connection instead of ten loose exports.
3. Performance always at hand. Weekly and monthly figures per channel, including comparisons with last month and last year, directly available as context.
4. Reports straight from the source. No templates, but documents about your account, fed from your own knowledge layer.

The difference with "just using ChatGPT" is the same as the difference between a new employee you have to onboard again every week and an experienced colleague who already knows everything.

Control restored: a human approves everything

And then the second problem: the grip. For me, this is the core.

I don't believe in autonomous agents tinkering with campaigns and systems unseen. I believe in experienced people checking AI output before anything happens. The AI is lightning-fast at the groundwork; a human reviews the result and decides. That is why we build every MCP with four fixed safeguards:

1. Permissions per role. You determine in advance which data the AI can see and which actions are possible at all. Per user, per system. The AI only sees what it is allowed to see.
2. Approval first. The AI prepares optimisations, in Google Ads, in Meta, wherever. But it never executes anything on its own. Every proposal waits for a human.
3. Everything traceable. Every question, every piece of advice and every approval is logged. An owner can always look back at what happened, based on what, and who signed off on it.
4. Your knowledge stays yours. The brain runs in your environment, on your data. A specialist leaves or you switch tools? The context stays put.

For a marketing manager this means: your team works faster, but within boundaries you set. For an owner it means: you know what AI is doing in your company, and innovation no longer means handing over control.

Not a pilot, but a proven approach

This is not a whiteboard story. The MCP architecture behind MAXai was developed and proven in production: the platform runs for more than 25 clients. Client brain, semantic search in your own knowledge, performance snapshots, automatic reports and prepared optimisations, with a human at the wheel at every decision point.

We now build that same foundation custom. For marketing teams that want more than loose AI experiments. For owners who want to accelerate without losing their grip. And for agencies that want to deliver AI that really knows the client.

The question you should ask yourself

Not: "are we using AI?" Everyone is by now.

But: "do I know what our AI knows, and who decides?"

If you can't answer that question, you don't have an AI strategy but an AI gamble. The technology to fix that exists, is open and is proven. What's missing is the architecture: a knowledge layer that knows your business, where nothing happens without your approval.

That is what we build. And the keys? Those stay with you.