If this is one of the first articles you’re reading from our blog, I’m Tom, co-founder and CTO of Detrics. And this is the article where I tell you why we dove headfirst into a technology that’s less than two years old.
The short version
We found out Anthropic was releasing an open protocol to connect AI with external tools, and we said: let’s go all-in.
The long version is more interesting.
Detrics before MCP
Detrics is a data platform for digital marketing. We connect over 50 sources (Facebook Ads, Google Ads, TikTok, Shopify, Google Analytics, Stripe, and more) and give our users a way to query that data from Google Sheets, Looker Studio, or our web app.
We’ve spent years building the infrastructure that translates each platform’s API into a unified format. Every metric, every dimension, every filter. It’s thankless, detailed work that you only appreciate when it works, because when it works you simply see your data where you need it.
But we always had a frustration: no matter how much we simplified querying data, users still had to know what to ask and how to ask it. You had to know that “CPC” is called “cost_per_click” on one platform and “average_cpc” on another. You had to build the query yourself.
The “this is what was missing” moment
When Anthropic announced MCP, something clicked. It wasn’t just another API. It was a standard that said: “this is how AI connects to external tools, in a secure and structured way.”
And we already had all the data infrastructure built. What we were missing was the last mile: letting AI make the query on behalf of the user.
Instead of you sitting down to build a report, you tell Claude: “How much did I spend on Facebook this week and how does it compare to last week?” and Claude uses our MCP to go fetch the answer. Without you needing to know which fields to request, in what format, with what filters.
Adoption was fast. Slack, Notion, and GitHub all jumped on board. Today there are over 8,600 MCPs in the world, barely a year after launch. We weren’t alone in seeing the potential.
The decision to go all-in
Honestly, an MCP doesn’t even seem like that big a deal to me now. Technically it’s a relatively simple layer. But what it enables is enormous.
The potential is in putting AI into every process in a more integrated way. From wherever you happen to be working, your code editor or your chat window, you can reach your business data right there in the conversation. You stop switching context, opening another tab, and exporting a CSV to answer a question you could have just asked.
And for us at Detrics, it meant that all the data infrastructure we’d built over years suddenly had a completely new distribution channel. We no longer depended on users learning our interface. The interface was the conversation.
What we built
Our MCP connects Claude (and soon other models) to all the platforms we already support:
- Advertising: Facebook/Meta, Google Ads, TikTok, Pinterest, LinkedIn, Bing, Twitter/X, Amazon Ads
- E-commerce: Shopify, WooCommerce, MercadoLibre, TiendaNube
- Payments: Stripe
- Analytics: Google Analytics, Google Search Console
You can ask things like:
- “What are my top 5 campaigns by ROAS this month?”
- “Compare my Shopify sales from January vs February”
- “How much did I spend on ads this week, across all platforms?”
And the AI goes, queries the real, up-to-date data, and responds. In seconds.
Why Anthropic and not someone else
I’ll be direct here because this is more opinion than fact: Anthropic is a high-compute, high-intelligence-density company with heavy R&D investment. In our testing, Claude comes closest to understanding a complicated instruction and then executing it with some judgment about the edge cases.
Publishing MCP as an open standard was a smart move on their part. Instead of building a closed ecosystem, they opened the protocol so anyone could build on it, and the trust that bought them is why the ecosystem filled up so fast.
We’re not an AI company. We’re a data company. But we understand that the future of data is about how AI consumes it, not just a human looking at a chart.
What we learned
Building an MCP taught us something we already suspected: the hardest part isn’t the technical connection. The hardest part is the quality of context.
AI is only as good as the information it receives. If you pass it raw, unstructured data, it’ll give you generic answers. If you pass it well-named fields with clear descriptions and correct units, the AI understands what it’s looking at and gives you answers that actually help.
That’s why at Detrics we don’t stop at connecting data. We translate it. Every field has a normalized name, a description, a category. When AI queries our MCP, it knows that “spend” on Facebook is conceptually the same as “cost_micros” on Google Ads (divided by a million). Most of the work we’ve done sits in that translation layer.
What’s next
Today we’re focused on the most popular data sources for digital businesses. The direction is clear though: we want to be the data layer that gives any AI its business context.
Connecting more platforms is the easy half. The harder ambition is an AI that understands your business end to end. Crossing your ad data with your sales, your sales with your inventory, your inventory with your finances. All in one conversation.
We already run our own operation this way. We use AI to write code, analyze data, and design features, and the honest result is that the team does more of the thinking and less of the fetching. That’s the direction we want the product to push people in.
MCP is one of those tools, and it’s still very early.
If you want to try the Detrics MCP, you can set it up in 2 minutes with Claude Desktop or Claude Code.
If you’re interested in how this changes reporting, read: Why MCPs Are the Future of Reporting →
