Neural Concept SA, a Switzerland-based engineering company built around AI, has launched Model Context Protocol Bridge, a capability that lets AI agents operate commercial computer-aided design and simulation software. As first reported by SiliconANGLE, the aim is to help large engineering organizations automate some of the most complex software they use to design and test physical products. The Model Context Protocol is the connective layer here, and its arrival in CAD and CAE is worth a close look.
Most agent announcements focus on code, customer service or IT operations. This one targets a different kind of work: the slow, detail-heavy preparation that happens before a car part or an aircraft component ever gets simulated.
What the Model Context Protocol Bridge does
Neural Concept says the bridge gives AI agents access to the engineering tools they need, along with the domain know-how to use them properly. The company's platform already folds deep learning into CAD and CAE workflows. The Model Context Protocol Bridge extends that by letting an agent act inside the software engineers already rely on, rather than sitting beside it.
Ownership is a key part of the pitch. According to the company, customers keep control of their software and their data. For manufacturers in defense, aerospace and semiconductors, which are all among Neural Concept's customer base, that condition is likely to decide whether a pilot gets approved at all.
Chief Executive and co-founder Pierre Baqué framed the launch as a first: software that shapes every physical product can now be operated by AI at enterprise scale. That's a big claim, and the company hasn't published broader evidence to back it up. The customer examples are more concrete, and we'll get to them below.
Why simulation prep is the bottleneck
Simulation is how engineers prove a design before building physical prototypes. Running the solver, the step that actually crunches the physics, is already largely automated. The slow part comes before it: pulling the right geometry, cleaning it up and turning it into a mesh a simulation can use.
That preparation is what makes a digital twin useful. A twin can only be simulated in high fidelity if it behaves and looks like its real-world counterpart, which means working from spec sheets, materials guidelines and a large volume of data. The work is tedious and needs extensive validation.
The payoff for doing it well is considerable. Once a model is ready, engineers can test thousands of design scenarios at once in virtual labs instead of waiting on physical builds. But getting a model to a rough draft that an expert can refine can take weeks or longer. Neural Concept says an agent working inside CAE software can bring that down to hours, with the engineer still doing the final tuning.
Early results from customers
The company points to two data points. In one customer case, an engineer used the bridge to clean up the geometry of a wheel rim in three minutes. The same job done manually took a day.
The second comes from Ciro Alosa, an electromagnetic design engineer at Traton SE, the Volkswagen Group subsidiary that oversees its heavy truck, bus and light commercial vehicle brands. He said agents have changed how he works. He can run more tasks in parallel, and moving from a research paper to a validated script in his team's simulation toolchain now takes a couple of days.
Both are single examples, not benchmarks, and neither tells us how the Model Context Protocol Bridge performs across a full program. Still, they point to where the gains show up: repetitive preparation and scripting, not the judgment calls at the end of the process.
Why the Model Context Protocol matters for engineering software
The Model Context Protocol is generally used to let AI agents connect to outside tools and data in a consistent way. Commercial CAD and simulation packages are some of the most intricate tools an agent could be asked to drive. They have deep menus, strict file requirements and results that need verification. A bridge built on the Model Context Protocol gives agents a defined way in, instead of relying on fragile screen-level automation.
There's also a staffing angle. Baqué noted that manufacturers want shorter development cycles while fewer experienced engineers are available. If an agent can handle an approved task, the hours it spends are hours an expert gets back for harder problems. That's a more modest promise than replacing engineers, and probably a more believable one.
What to watch next
For technology leaders at manufacturers weighing a Model Context Protocol rollout, a few questions will matter more than the launch itself:
- Scope of approved tasks. How much of the preparation work can be handed to an agent before a human has to sign off?
- Validation. Simulation results drive real manufacturing decisions, so teams will want a clear way to check what an agent produced.
- Data control. The company says customers keep ownership of software and data, and buyers will want to see how that works in practice.
- Breadth of tools. Large engineering shops run many commercial packages, so coverage across them will shape real adoption.
Neural Concept hasn't said how widely the bridge is deployed, and the figures so far come from the company and its customers. Even so, the direction is clear. Engineering software is becoming something agents can operate under human supervision, and the Model Context Protocol looks set to be the way many of them get in. Teams that spend weeks on model preparation should be watching how this plays out.
Source: SiliconANGLE
