Even the most sophisticated AI (artificial intelligence) models are constrained by data trapped in information silos and legacy systems. The challenge is every new data source requires its own custom implementation, making connected systems difficult to scale, which is something that could slow building the next-gen infrastructure projects we so desperately need. Enter MCP (model context protocol).
In November 2024, Anthropic introduced the model context protocol, which is a standard for connecting AI assistants to the systems where data lives. Anthropic suggests MCP provides a universal, open standard for connecting AI systems with data sources. The objective is to replace fragmented integrations with a single protocol.
Francois Valois, senior vice president, Bentley Open Applications, Bentley Systems, says model context protocol is an open model agnostic standard that connects several types of technology and engines to the large language model technology.
“It is unlocking a lot of value on the agentic workflow,” he says. “It is the agentic foundation, providing the deterministic answer, so the right answer, the 99.9%/100% accurate answer, to nondeterministic generative AI such as ChatGPT or Claude and so on. It is really connecting these two worlds and exposing the engineering engines to these smart, generative AI tools.”
Why is this so important when we are talking about our critical infrastructure? With our roads, bridges, water, energy, aviation, you name it, we want structures to be built accurately.
In a recent blog post, Julien Moutte, chief technology officer, Bentley Systems, so aptly says, “A model that is 90% right is a useful start; a structural analysis that is less than 100% right is a catastrophic liability. Would you drive across a bridge that is ‘hopefully’ designed right?”
MCP Building Blocks
In my exclusive conversation with Valois, he explains MCP is one of the building blocks that is evolving and making this possible. “It isn’t the answer to all our problems, but it is one of the tools in our toolbox.” A much-needed tool, I might add.
This all leads to the power of openness, a core position of Bentley Systems, a conversation I have had many times with Moutte. Moutte is passionate about open systems. In fact, he has stated if data doesn’t flow openly and consistently from design through operations, owners just might be paying for the same information repeatedly, which could end up costing them millions.
For Bentley Systems, openness is at its core. As Valois explains, it has spent years building the iTwin platform, and its 2024 acquisition of 3D geospatial leader Cesium only accelerates that vision. The combination of iTwin and Cesium gives developers the ability to align 3D geospatial data with engineering models, subsurface information, IoT (Internet of Things) streams, reality capture, and enterprise data, the full stack needed to power true digital twins.
And this truly gives designers what they want. They can begin their work using real world context, capture field data, survey scans, point cloud, LiDAR, and more instead of just recreating existing conditions all over again. The result is faster starts, improved accuracy, and designs that are grounded in reality from the beginning.
Bentley Systems recognizes it’s part of a larger ecosystem. Design may last a year, then construction a few years after that, but the operations will last for decades. Owners and operators need to be able to write a program that can still access that data decades from now.
“MCP is part of that strategy,” Valois says. “Not only are we opening our data, but we are also opening our tools to the rest of the world.”
Recently, Bentley Systems released MCPs for MicroStation and STAAD. Bentley uses MCP to connect AI agents directly to validated engineering tools like STAAD, which is built for real world structural engineering. In the future, Bentley Systems is going to build MCPs for all of its open applications and other tools as well.
The Future of MCP
As we look to the future, the relationship between engineers and AI is evolving. Valois is clear AI is augmenting the engineer and not replacing them. As we move forward into this new era of work, he suggests the first step is to understand what AI can do and understand its limitations as well.
Ultimately, engineers need to be in control of the technology. This is true for everyone from big firms to small firms and from contractors to owners. Valois suggests starting with the outcome you need and then beginning with a small project. Then repeat as needed.
For infrastructure, MCP represents a framework and an opportunity to create a common language to help construction professionals unlock valuable insights without sacrificing accuracy. MCP’s greatest impact will be enabling engineers to spend less time managing AI and more time making informed decisions about a project, which is key when building assets that must perform for decades.
“It’s a beautiful technology because it is so simple and that is why it took the market by storm,” says Valois.
Want to tweet about this article? Use hashtags #construction #IoT #sustainability #AI #5G #cloud #edge #futureofwork #infrastructure #MCP #open #STAAD