NOV 18, 2026
Bentley Opens MicroStation AI Link to All Users of Latest Release After Strong Early-Access Demand
Company pulls the tool out of a limited trial months ahead of its planned commercial release as AI adoption spreads across engineering firms.
By Tomas Kellner
Key Takeaways
- MicroStation MCP is now available to all users of the latest MicroStation, months ahead of schedule, thanks to early-access demand.
- The tool lets any MCP-capable AI assistant work inside a live MicroStation session from plain-English instructions.
- Bentley’s Stuart Milne twice used the tool to re-create London’s iconic Gherkin—with Claude and Astra—cutting work that once took days to just 30 minutes, with the engineer still checking every result.
Bentley Systems has released the AI connection for its MicroStation design software to every user of the software’s latest release, pulling the technology out of a limited trial months earlier than planned after high demand from engineers.
The tool, called MicroStation MCP, lets an AI assistant plug into a live MicroStation session. Engineers type commands in plain English—prompts like “analyze a design,” “check a drawing against standards” or “automate a repetitive modeling task”—and the AI completes the task inside the software. After an automated modeling task, the AI also helps users check its own work for accuracy and inconsistencies. The tool works with any AI agent that supports the Model Context Protocol (MCP), so users can keep the AI assistant they already use.
Bentley had planned to commercially ship the feature with MicroStation 2027. But with so many sign-ups for its Early Access Program, the company opened it to all MicroStation 2026.2 users. Bentley said the move reflects the speed with which engineers are adopting AI.
Bentley began rolling out MCP servers across its product line in the spring. At the company’s Year in Infrastructure Awards in Singapore in October, CEO Nicholas Cumins said close to 70% of the 2026 finalists used AI, but the standout teams had to write their own code to do it, something most engineering firms can’t do.
“What if engineers could simply interact with AI in natural language, and let AI translate their intent into software instructions on the fly?” Cumins said. “That is what is now possible.”
Few users have pushed the tool harder than Stuart Milne. Two decades ago, Milne was the CAD coordinator at the leading global architecture firm Foster + Partners. The company’s office was in 30 St Mary Axe, the London tower that locals have since nicknamed the Gherkin for its elongated curves and green color. At the time, Foster + Partners used an early MicroStation feature called dimension-driven design, one of the first forms of computational design, together with Visual Basic for Applications (VBA). Teams used the system to generate the tower’s trademark curved form and twisting facade from thousands of triangular panels, one panel at a time. When the building was honored at the first Bentley YII Awards in 2004, the project submission said models that once took days to draft could be generated in minutes.
In June, Milne—now a product manager for MicroStation at Bentley in Sydney, Australia—wanted to try again, with new technology. He loaded everything he could find about the tower into Google’s NotebookLM, connected that and MicroStation to Anthropic’s Claude through MCP, and let the AI rebuild the shell of the building using the same formula that Foster + Partners’ team used in 2004. The AI system placed roughly 9,000 parametrically driven panels, whose shapes are generated by rules and dimensions rather than drawn one by one. Each was a different shape and size, and each floor of panels was slightly different from the one before.
It worked, but Milne had to spell out every step. “I had to talk in more CAD language what I wanted the CAD system to do,” he said.
In September, he tried again with Astra, then OpenAI’s latest public model, and went after the building’s entrance lobby, where the tower’s neat repetition breaks down. He provided context in the form of reference images, website links, pattern instructions, and text descriptions of the entrance and the atrium. This time, he said, “I could talk to the model as if it was a building rather than in CAD language.”
Another Bentley colleague then used AI connected to the MicroStation model to generate the structure in STAAD, Bentley’s structural-analysis software. It read the model and placed structural nodes in the right spots.
In one demonstration, Milne asked the AI to unwrap each of the lobby’s column casings into a flat matrix of pieces, ready to be cut and folded off site. “It came back and did the job in about half an hour, completing the task, and provided a reusable script that could be used again and again, should the design change,” Milne said. “This manual process could take days each time, or even writing a manual script could take days by hand.”
Milne said the reusable script is important. It means that “AI is not needed for future changes,” he said. “It just runs the script. This means time and cost are reduced, and determinism becomes certain, by taking AI out of the loop for future runs.”
Still, Milne isn’t ready to hand over his work to AI anytime soon. “It still makes assumptions that look plausible but aren’t quite right,” he said. “The user still needs to guide the model, correct the model, check the model.” Bentley CEO Cumins put it more bluntly in Singapore: “An algorithm does not sign and stamp the drawings.”
The new AI connection for MicroStation arrives through a Bentley Copilot update inside MicroStation 2026.2 and isn’t available in earlier releases.
Milne doesn’t think the tool makes engineers less necessary. “Those years of knowledge, being able to translate what your end goal is into words and building the context around that, becomes even more valuable,” he said.

