IndustriConnect: MCP Adapters and Mock-First Evaluation for AI-Assisted Industrial Operations
New research bridges AI and factory floors with 870 test runs across Modbus, MQTT, and OPC UA protocols.
A team of researchers led by Melwin Xavier has published a paper on arXiv detailing IndustriConnect, a prototype system designed to bridge the gap between modern AI assistants and legacy industrial infrastructure. The core innovation is a suite of Model Context Protocol (MCP) adapters. MCP is a standard, popularized by Anthropic, that allows AI models to discover and use external tools. These adapters translate complex, safety-critical industrial communication protocols—specifically Modbus, MQTT/Sparkplug B, and OPC UA—into a schema that AI agents can understand and safely operate. This means an AI could, in theory, be tasked with a multi-step workflow like "monitor sensor X and adjust valve Y" without needing to be explicitly programmed for the underlying machine communication.
Crucially, the system employs a 'mock-first' evaluation workflow, allowing developers to rigorously test adapter behavior locally with simulated equipment before ever connecting to real, potentially dangerous plant machinery. The team conducted a massive deterministic benchmark comprising 870 total runs (480 normal, 210 fault-injected, 120 stress, 60 recovery). Results showed the adapters achieved 100% success in normal scenarios, correctly handled structured errors (like out-of-range uint16 values), identified concurrency limits under stress, and successfully recovered sessions after simulated endpoint restarts for all three protocols. This provides concrete evidence for the correctness and safety of AI-assisted industrial operations.
- Bridges AI and industry: Creates MCP adapters for Modbus, MQTT, and OPC UA, turning industrial controls into discoverable AI tools.
- Mock-first safety: Enables 870 benchmark runs (including fault and stress tests) in a simulated environment before touching real equipment.
- Proven reliability: Normal operations achieved 100% success; system demonstrated structured error handling and session recovery across all protocols.
Why It Matters
This research paves the way for applying powerful AI workflow automation to manufacturing, energy, and logistics while prioritizing critical safety controls.