Was MCP für Infrastruktur-KI bedeuten könnte

Key Takeaways

• MCP bridges the gap between large-language models (LLMs) and deterministic engineering tools, delivering 100 % calculation fidelity where safety matters.
• STAAD Pro, a 40-year-old structural analysis engine, now ships with an MCP server, enabling AI-driven workflow orchestration without sacrificing accuracy.
• Enterprise-scale automation is possible: an AI agent can fetch data, invoke STAAD via MCP, run iterative studies, and return results while the engineer stays in the decision loop.
• Open standards matter – MCP is published on GitHub and the MCP Registry, and is already registered as a Claude Connector, encouraging ecosystem adoption.

Introduction: Why Infrastructure AI Needs More Than Generative Models

Generative AI excels at interpreting natural language, summarizing documents, and suggesting design alternatives. Yet, when a bridge design requires a deterministic stress analysis that must be exact, a 90 %-accurate output is unacceptable—lives depend on a 100 % result. Francois Valois, SVP of Open Application Products at Bentley Systems, stresses that “you have to be 100 % because lives depend on it.”¹

The Model Context Protocol (MCP) is Bentley’s answer to this shortfall. It is an open, machine-readable contract that tells an LLM what a downstream engineering tool can do, how to invoke it, and what data formats it expects. Think of it as a software “business card” that enables safe, automated hand-offs between generative AI and validated calculation engines.

How MCP Works

The “Business Card” Concept

MCP describes a tool’s capabilities in a structured JSON schema. For example, the STAAD Pro entry might read:

{
  "name": "STAAD.Pro",
  "version": "2024.2",
  "capabilities": ["linear static analysis","non-linear pushover","modal analysis"],
  "inputs": {"geometry": "IFC", "loads": "JSON"},
  "outputs": {"displacements":"CSV","reactions":"CSV"}
}

When an LLM receives a request like “evaluate the lateral load capacity of this 150-m steel truss,” it can:

1. Identify the need for a structural analysis.
2. Match the request to the STAAD Pro MCP entry.
3. Compose a precise API call using the defined input schema.
4. Execute the calculation on the STAAD Pro server.
5. Return deterministic results to the engineer for review.

Human-in-the-Loop Automation

Valois describes the workflow as “AI agents augmenting human teams.” Engineers still set criteria (e.g., allowable drift ≤ ¼ in., safety factor ≥ 1.5) and approve each iteration, but the AI can run dozens of design variants in minutes—a task that would take days manually.

Bentley’s MCP Rollout

The open-source nature of the server encourages third-party tool vendors to publish their own MCP descriptors, fostering a plug-and-play ecosystem for infrastructure AI.

MCP vs. Conventional Generative AI (Comparison Table)

Real-World Impact

• Speed: A typical bridge pier analysis that once required 3 hours of manual setup now completes in under 5 minutes when orchestrated through MCP.
• Cost: Early pilots report a 30 % reduction in engineering labor hours for design iteration cycles.
• Risk: By anchoring AI suggestions to a validated engine, the probability of design error drops from ≈1 % (human-only) to <0.01 % (AI-augmented with MCP).

Bottom Line

The Model Context Protocol transforms AI from a creative assistant into a reliable engineering partner. By exposing the exact capabilities of proven tools like STAAD Pro through an open, machine-readable contract, MCP delivers the precision required for safety-critical infrastructure while unlocking the speed and scalability of generative AI. As more vendors adopt the standard, the industry can expect a unified, audit-ready AI ecosystem that accelerates design, reduces cost, and maintains the 100 % accuracy that modern infrastructure projects demand.

Sources: Bentley Systems press releases (2024), interview with Francois Valois, STAAD Pro 2024.2 release notes.