Everyone knows what a game-changer AI can be. Give it a prompt, and it can write, analyze data, or automate entire workflows. But until recently, you had to feed it a lot of context to get useful results.
But that’s starting to change with Model Context Protocol (MCP)—an AI standard that connects external solutions so that tools can talk to each other more effectively.
In this episode of Digital Builder, I sat down with Ben Cochran, VP of Engineering at Autodesk, to unpack all things MCP and how construction pros can leverage it. Check out our conversation below.
We discuss:
MCPs are protocols that enable AI agents to connect directly with external tools and data sources. Ben likens them to APIs, but for large language models (LLMs).
“Think of it as an API for an LLM,” he explains. “What a protocol does is it helps build standards so that AI agents can have richer experiences, where they can tie into data sources without having to learn what the API spec is for that particular data source.”
APIs are a useful comparison point because, like MCPs, they help connect different applications and systems. But it’s important to note that there are clear distinctions between MCPs and APIs.
As Ben explains, “An API is a broad term. It’s an interface that lets an application connect to another system, often through a REST protocol. The LLM, though, has to figure out what that API looks like, read the documentation, and build a connection.”
MCPs, on the other hand, remove that manual work. “Agents are like us; when they have a bunch of work to do and you remove that work, they can get onto the next thing,” says Ben. “An MCP helps accelerate the opportunity to access that data source.”
Ben adds that APIs still play an essential role in the age of MCPs and AI-driven workflows.
“APIs are still critical, even more critical with MCPs and with agent workflows and the richness that you get from LLMs,” he says. “APIs allow systems to connect with each other. You’ve got applications that come from Autodesk and its services and systems, and then you’ve got third-party tools that build out an entire ecosystem. Those are all connected together with APIs.”
MCPs have grown in popularity over the last few months, mainly because of their ability to expand what AI agents (also a popular topic) can do.
Ben says it best: “It’s a rocket ship built on a rocket ship.”
LLMs are already transforming how we work. We use it to summarize emails, analyze data, and automate repetitive tasks. But for a while, those capabilities were contained. “You were either using a tool, or you were using this agent, like Claude or Copilot,” says Ben. In other words, you had to jump from system to system to get things done.
What’s fueling all the excitement around Model Context Protocols is that they break down those silos. “Now what you’re doing is you’re asking the model to do an action,” Ben explains. “You’re actually describing an outcome that you want to complete, and then that agent is now operating almost like an intern would.”
He adds, “I like to refer to them as interns—agents as interns—because they’re very capable, they reason very well, but they don’t have any context for what they’re trying to do. That’s our job: to set that context and give them the outcome or result of what they’re trying to do.”
Until now, that “intern” was limited to a single tool. But with MCPs, the walls come down. “Now I can have other systems like Autodesk Construction Cloud or Revit, or maybe a SAP database,” Ben says. “All of a sudden, I can connect to that other data source and pull that data in, and now the agents have a far wider scope that they can operate in.”
When you look at how construction teams work today, there’s no shortage of tasks that eat up time and energy. So, one thing that gets Ben excited about MCPs is their potential to flip that dynamic entirely.
“There are a bunch of things that we compromise on because they’re either too time-intensive or they cost too much,” says Ben. “And those are usually related to people—people’s time and people’s costs.”
With MCPs, those constraints start to loosen. Instead of weeks, certain processes can take moments. Teams can run more design alternatives, audit documents faster, and prepare project data for bidding or validation—all without the repetitive manual work that typically slows things down.
One use case? BIM validation.
“Right now, the way you do this is you write a bunch of code that says, ‘these are the valid objects in my document, and these are invalid,’” explains Ben. “Then you flag them, and that creates a bunch of tasks for the engineer who has to go clean up the document. That’s a lot of work just to go through that list.”
“Now the engineer is reviewing the changes instead of actually having to look up what all those changes are,” Ben says.
If you’re intrigued by MCPs but aren’t sure where to begin, Ben says it’s best to start small and build from there.
“For one, I would look at and actually start with agentic tools—ChatGPT, Claude, Perplexity. I would grab these tools, there’s a bunch more, install them, work with them, and try to understand how they can help me.”
The goal, he says, is to practice how you prompt. “I end up writing more and more actually, at the prompt than I started with when I first began,” Ben says.
“I used to ask a question like I was Googling something. Now I go in and say, I need this—and give a whole rundown of everything I need, everything I want. I’ll even go, if I have an example of something, I’ll say, I want it to be in this format using this example.”
Once you’re comfortable with agentic tools, try connecting one to a data source using an MCP. “It’s really just a URL,” Ben explains. “You take it, and you say, I want to connect to this service or get data from this provider. If they have an MCP, you plug it in—and now your agent can go get that data and pull it into your content.”
Digital Builder is hosted by me, Eric Thomas. Remember, new episodes of Digital Builder go live every week. Listen to the Digital Builder Podcast on:
or wherever you listen to podcasts.
