What MCP Means for Learning Data (and Why It Matters Now)

There’s a lot of noise right now around AI in learning.

Most of it focuses on generation. This means faster content creation, smarter chat interfaces, automated workflows. All useful, to a point. There’s a quieter shift happening underneath all of that. One that has less to do with what AI creates and more to do with what it can access.

That shift is happening through something called Model Context Protocol (MCP).

It’s not a flashy term, and it doesn’t demo like a chatbot. But it may end up being one of the more important changes in how learning systems actually work.

From “Answers” to “Context”

Most AI in learning today operates in a constrained environment. You give it a prompt, maybe some uploaded content, and it produces an answer.

Sometimes that answer is useful. Sometimes it’s confidently wrong. Either way, it’s operating with limited context.

MCP changes that dynamic.

Instead of treating data as something that has to be preloaded or manually integrated, MCP allows AI systems to reach out, securely and in a structured way, to the systems that already hold your operational truth. Your LMS. Your LRS. Your SOP repositories. Your HR systems.

The difference is subtle but important.

The AI is no longer guessing based on what it was trained on. It’s grounding its responses in what is actually happening inside your organization by accessing data where it lives.

A Practical Example: The Veracity LRS MCP Server

We’ve been exploring this directly through our work on the Veracity LRS MCP Server. In this setup, the AI agent isn’t working from static exports or prebuilt dashboards. It’s connected to a live stream of xAPI data consisting of learning activity, interactions, behaviors, and it’s structured in a way it can query and interpret in real time.

That changes the nature of the interaction.

Instead of asking, “What does this report say?” you can ask, “What’s happening?”, and get an answer that’s grounded in actual data, not a snapshot from last week.

Why This Matters for Learning Measurement

For years, learning measurement has been constrained by tooling.

Even in organizations that invested in xAPI and an LRS, the reality often looks like this: the data exists, but accessing it requires effort. Queries have to be written. Reports have to be built. Dashboards have to be maintained.

So what happens? People default to what’s easy. Completion rates. Quiz scores. Surface-level metrics that are readily available but rarely meaningful.

MCP doesn’t magically fix measurement. But it removes a major barrier.

It makes the data more accessible. This isn’t just for developers or business analysts, but for the people who are actually trying to make decisions.

That opens the door to better questions. And better questions tend to lead to better outcomes.

What This Looks Like in Practice

Consider a fairly common scenario.

An LMS administrator notices that a group of new hires isn’t performing at the level expected. In a traditional setup, that might trigger a familiar workflow: pull completion reports, export data, try to correlate it with whatever performance metrics are available, and hope something stands out.

With an MCP-enabled agent connected to the LRS and relevant systems, the interaction becomes much more direct.

You can ask: “What learning activities have this group completed in the last 30 days, and how does that relate to their early performance indicators?”

The agent retrieves the data, identifies patterns, and surfaces insights without requiring someone to stitch together multiple reports. The friction between question and answer is dramatically reduced.

A different kind of question emerges when you start thinking about content usage.

Most organizations don’t actually know which resources are helping people in the flow of work. They know what’s assigned. They know what’s completed. But those are not the same thing.

With MCP, you can shift the lens.

Instead of asking what was delivered, you can ask: “What resources are people actually accessing while they’re doing their jobs?”

Now you’re looking at behavior in context. You’re starting to see which assets are truly useful and which ones are just… there.

And then there’s the operational side.

One of the more interesting capabilities MCP introduces is the ability to move from one-off analysis to ongoing awareness. Instead of manually checking reports, you can set up routines in your AI tools that do that for you.

For example: every week, summarize learning activity related to safety procedures and flag anything that looks unusual.

Now the system is watching with you. Not replacing judgment, but augmenting it. Bringing forward patterns that might otherwise go unnoticed.

A Shift for System Builders

If you’re building learning platforms or internal systems, MCP changes the design conversation.

For a long time, the expectation has been that software should anticipate user needs by providing dashboards, reports, and predefined views into data. The result is often a growing collection of features that are expensive to build and maintain, and still don’t quite answer the questions people actually have.

MCP offers a different approach.

Instead of trying to predict every question, you expose structured access to the data and let an agent handle exploration. The focus shifts from building interfaces to building a clean, reliable data layer.

It also makes it easier to connect learning data with other systems. HR, CRM, operational platforms—data that has traditionally lived in separate silos can now be brought together in a more fluid way.

That’s where some of the more meaningful insights start to emerge. Not just what people learned, but how that learning connects to performance, retention, and outcomes over time.

Where This Is Headed

This isn’t a fully mature ecosystem yet.

Most organizations are still early in their use of AI, let alone MCP. But the trajectory is becoming clearer. The focus is shifting away from static reporting and toward dynamic, contextual access to data. AI, in this model, acts less like a content generator and more like a bridge—connecting systems, retrieving information, and helping people make sense of it.

Of course, none of this works without good data.

If your underlying data is inconsistent or poorly structured, MCP just gives you faster access to bad inputs. But if you’ve invested in clean, well-structured data—especially through xAPI and an LRS—the upside is significant.

Installing and configuring MCp for your SaaS products is also still really really new. You’ll likely need a contact or tech support lift from your vendor or internal teams. You’ll need to monitor the MCPs performance and accuracy at first as well just to make sure you aren’t getting bad data or using up all your AI usage tokens needlessly. 

Final Thought

For a long time, learning teams have been sitting on data they couldn’t fully use. Not because it wasn’t valuable, but because it was difficult to access, interpret, and apply.

MCP starts to change that.

It shortens the distance between having data and doing something with it. And in a field that has historically struggled to connect learning activity to real-world outcomes, that’s a meaningful step forward.

It’s really not about making AI more impressive. It’s about making systems more useful.