This study proposes a multi-layered expert system architecture designed to optimize domain-specific decision support through four synergistic mechanisms. First, the implementation of the MCP (Model Context Protocol) standardizes data access, enabling seamless interfacing between MCP clients and heterogeneous data sources—including SQL, vector, and file-based databases—via predefined toolsets. Second, a specialized RAG (Retrieval-Augmented Generation) Engine facilitates complex query resolution by deconstructing tasks based on expert decision logic retrieved from a Vector DB; sub-tasks are then executed through either semantic vector retrieval or a T2SQL (Text to SQL) Engine for precise structured data extraction. Third, data integrity and security are maintained via a SCP (Secure Copy Protocol) Client for the transmission of proprietary domain documents. Fourth, an API Requester extends the system's knowledge boundaries through external API integration. By leveraging on-premise open-source LLMs refined through domain-specific fine-tuning, the proposed architecture achieves superior precision in analytical reasoning while effectively mitigating the significant operational costs and data privacy concerns associated with centralized cloud-based models.