Abstract

Large language models hallucinate. When deploying LLMs in an enterprise setting where decisions rely on precise and auditable data, their tendency to produce convincing yet false outputs turn hallucinations from a minor issue into a significant failure. The Model Context Protocol (MCP) and Retrieval Augmented Generation (RAG) are architectural solutions studied to address this problem, but most literature consists of theoretical frameworks and proof-of-concept proposals. This study develops and empirically tests a controlled, agentic MCP-mediated RAG system against a baseline language model. An agentic MCP-RAG system pairing Claude Sonnet 4.6 with a ChromaDB vector database was evaluated against a baseline condition using the same language model without the retrieval infrastructure. This study finds that while both systems are capable of answering accurately, the difference lies in what they can prove. The Wilcoxon signed-rank test confirmed that the MCP-RAG architecture produced a statistically significant improvement in citation traceability. This architecture enabled the language model to retrieve evidence, cite its sources, and provide enterprises with a verifiable provenance chain that meets the accountability standards of regulated sectors. This is a capability that extends beyond a marginal improvement that previously did not exist due to the structural configuration of the baseline. The agentic MCP-RAG server, with its integrated vector database, structured citation protocol, and cross-platform orchestration across Google Workspace and Notion, demonstrates that the gap between what standard language models can do and what enterprises need them to do is an architecture problem, and it is now solved.

Advisor

Ma, Changzhi

Department

Statistical and Data Sciences

Disciplines

Business Analytics | Business Intelligence | Computer Engineering | Technology and Innovation

Keywords

Model Context Protocol, Retrieval Augmented Generation, Agentic AI

Publication Date

2026

Degree Granted

Bachelor of Arts

Document Type

Senior Independent Study Thesis

Download

Share

COinS
 

© Copyright 2026 Saugat Shakya