MCP-Driven Proactive AAS: LLM-Assisted Process Modeling and Execution

Abstract

The digital transformation toward Industry 4.0 (I4.0) relies on the semantic interoperability of distributed assets, typically represented by the Asset Administration Shell (AAS). However, a significant gap remains in modeling and executing complex processes that adapt to dynamic environments and include humans as first-class participants. This paper introduces an agent-based framework that bridges this gap by integrating Large Language Models (LLMs) with a Model Context Protocol (MCP) infrastructure hosted within a proactive AAS Human-Machine Interface (HMI). The methodology transitions from manual discovery and modeling to an agent-guided cycle. The framework leverages LLMs to autonomously perform resource discovery using MCP tools, synthesize executable Business Process Model and Notation (BPMN) 2.0 process models through semantic reasoning, and facilitate decentralized choreography negotiation among independent AAS participants. An execution phase follows, utilizing a custom process engine that provides context-aware assistance to human operators during runtime. The framework is evaluated through a highly mechanized harvesting process in a Forestry 4.0 scenario. The qualitative results indicate that integrating MCP tools with detailed AAS semantics helps the user model technical AAS-based processes using BPMN 2.0 while providing real-time assistance for proactive interaction with AAS. By grounding agent-based behavior in standardized AAS action primitives, the approach offers a scalable solution for expressive, interoperable, and autonomous industrial workflows.