An Event-Driven AI Platform for Intelligent Business Process Automation with LLM and MCP Integration

Abstract

Modern enterprises face growing complexity in managing corporate infrastructure due to fragmented automation scripts, limited AI integration, and the absence of standardized interfaces for connecting language models to enterprise tools. This paper presents the architecture and prototype implementation of an event-driven AI platform for intelligent business process automation. The platform integrates Apache Kafka as a central event bus, large language models (LLMs) through the Dify orchestration layer, and the Model Context Protocol (MCP) for standardized tool access. A modular service abstraction comprising Source services for event ingestion, Core services for outbound operations, and user automation services implementing business logic enables developers to focus on domain logic while the platform handles infrastructure concerns. The system was deployed in a production corporate environment with more than 160 active automation services, 175 Kafka topics, integrations with more than 11 enterprise systems, and approximately 2,000 daily users. Integration of Retrieval-Augmented Generation (RAG) and MCP-based AI agents enabled conversational automation scenarios. In one evaluated scenario, task completion time was reduced from approximately 10 minutes to less than 2 minutes, corresponding to an 87% reduction.