Reporting, by which I mean any task that involves looking at data searching for meaning. And if you work in digital marketing, e-commerce, or any online business, you know that task takes up a big chunk of your week.
You open a dashboard. Look at numbers. Try to understand what happened. Cross-reference with another dashboard. You’re missing a data point. Go look for it on another platform. Build a spreadsheet. Make a table. Send it to your boss. Your boss asks about something that wasn’t in the table. Start over.
MCPs are going to break that cycle. And I’m not saying this as a vague prediction; it’s already happening.
Reporting was never short on data
Reporting doesn’t suffer from a lack of data. It suffers from the opposite: too much of it, scattered across too many places. Facebook Ads has its metrics. Google Analytics has its own. Shopify another set. Each with its own interface, its own logic, its own export limitations.
And this doesn’t just happen to the data analyst. It happens to the online store owner who wants to know if their ad spend is working. It happens to the marketer who needs to justify this month’s budget. It happens to the designer who wants to understand which creatives are converting best. It happens to the copywriter who needs to know which messages resonate with the audience.
Everyone, at different levels, ends up doing the same mechanical work: go find data, format it, try to understand what it means. And most of the time goes into collection, not analysis.
And the dashboard, which was supposed to solve this, has its own problems. A dashboard is a static snapshot of metrics that someone decided were important. But the questions that come up day-to-day aren’t always on the dashboard. “Why did ROAS drop on Tuesday?” doesn’t have a widget. “Which audience is converting best this month vs. last?” doesn’t either.
What changes with MCPs
An MCP gives AI direct access to your data sources. Instead of you going to fetch the data, you ask the AI and it goes. That part is easy to picture, and it’s the smaller half of the change.
The bigger half: AI has a bigger context than you do.
I don’t mean that as an insult. It’s literal. When you look at a dashboard, your brain is processing 5, 10, maybe 20 metrics simultaneously. AI can hold hundreds of data points in its context, cross-reference them, and find patterns you miss. Not because you’re bad at it, but because it’s humanly impossible to sustain that level of context.
So MCP doesn’t replace reporting. It amplifies it. It gives you a copilot that has the same data as you, but with a different processing capacity.
The dashboard needs to have AI
An opinion I know will annoy people: in a few years the dashboard will be something you talk to, and the grid of charts will be the fallback view rather than the product.
Today the dashboard needs to have AI, plus the ability to go behind the dashboard. To ask questions about what’s underneath. Not just see that conversions dropped, but ask why and have the AI tell you: “Conversions dropped 15% this week. 80% of the decline comes from campaign X, specifically ad set Y that ran out of budget on Wednesday. The rest is a seasonal dip consistent with the same period last year.”
A dashboard can’t do that. An AI with access to your data via MCP can.
3 informed wrong decisions > 3 uninformed wrong decisions
Every business wants the shortest possible delay between something happening in its operation and someone finding out about it. That’s the premise this whole argument rests on.
Are you going to make incorrect decisions? Yes. We all do. But there’s a huge difference between getting it wrong because you didn’t have data, and getting it wrong having looked at all the available information. Informed decisions, even when incorrect, build knowledge. Blind decisions only build anxiety.
MCPs reduce that latency between “I have a question” and “I have a data-backed answer.” Not to zero, since we’re still in a refinement period. But from minutes or hours to seconds. And that changes the dynamics of how you operate a business.
What it means for a marketer
If today your reporting workflow is:
- Open 4 platforms
- Export data
- Build a spreadsheet
- Make charts
- Write conclusions
- Send a report
With a well-configured MCP, it becomes:
- Ask the AI: “Give me the weekly campaign performance report, compared to last week, with the top 3 anomalies”
- Review, adjust, send
You can do this today. Step 1 doesn’t completely disappear. You still need to know what to ask and validate the answers. But the mechanical work of fetching, formatting, and cross-referencing data is something the AI handles.
What it means for a founder or leader
If you’re a founder, CEO, or head of a department, your relationship with data is different. You don’t need to build the report. You need to understand in 2 minutes if things are going well or badly, and why.
Today that requires someone to prepare the information for you, or for you to sit down and interpret a dashboard that maybe didn’t have exactly the question you wanted to ask. With an MCP, you talk to the AI like you’d talk to your senior analyst: “How are we doing this month vs. last? Is there anything I should worry about?”
And the AI comes back with current data, the comparisons that matter, and whatever looks anomalous. The closest analogy is an analyst who happens to be awake at 6am on a Sunday when you get curious about last week’s numbers.
What about security?
A question that always comes up: “Is it safe to give AI access to my data?”
It’s a fair question. And the short answer is: it depends on who built the MCP.
The most important guardrail of an MCP is the distinction between read actions and destructive actions. In computing terms, it’s the difference between reading and summarizing your emails versus investing $100K in Bitcoin. That distinction is defined by the MCP builder, rather than by Anthropic, OpenAI, or the AI model itself.
At Detrics, our MCP is read-only. The AI can query your data, but it can’t modify campaigns, change budgets, or touch anything in your accounts. And it always asks for permission before each query.
At the protocol level, MCP runs on standards that have been in the field for a long time: HTTPS/TLS, which has been protecting internet communications for over 25 years, and OAuth 2.0, the authorization standard used by virtually every platform since 2012. The new part is a thin layer sitting on top of infrastructure that has already been attacked for decades.
The rule is simple: if you trust the platform you’re connecting to via MCP, trust that they’ll ask permission for every sensitive action and that your data won’t leak.
The near future
Today we’re in a phase where agentic models have gone from having one-on-one conversations with humans to having conversations with themselves in between. AI doesn’t just respond. It plans, executes, validates, and comes back with results.
Since November 2024, when Anthropic launched the MCP standard, the ecosystem has exploded. Today there are over 8,600 MCPs listed in public directories. In just over a year, we went from zero to an ecosystem where virtually any digital tool can connect with AI.
That means data convergence is going to accelerate. Today you can ask AI about your Facebook campaigns. Tomorrow you’ll be able to ask about your campaigns, your sales, your inventory, your finances, and your sales pipeline, all in the same conversation. AI will be able to cross-reference data that today lives in separate silos and give you a unified view of your business.
Sound like science fiction? Three years ago, asking an AI to write a persuasive email in regional Spanish was too.
Why it matters who builds the MCP
MCPs vary enormously in quality, and the AI’s answers vary with them. Feed a model badly formatted, incomplete or slow-arriving data and you get mediocre answers back, which most people then blame on the model.
At Detrics we work on exactly this: the data infrastructure connecting over 50 marketing and e-commerce platforms to AI models. Every field, every metric, every dimension is mapped, so the AI knows what it is looking at before it starts reasoning about it.
That mapping work is unglamorous and it is most of the job. A model can only be as good as the context it gets handed.
I’m Tom, co-founder and CTO of Detrics. If you want to try our MCP, you can set it up in 2 minutes.
If you’re interested in why we decided to go all-in on this technology, read: Why We Built an MCP at Detrics →
