ME-RAG presents a novel multi-agent architecture designed to enhance retrieval-augmented generation (RAG) through structured, role-based discussions. The system leverages agents with different perspectives, enabling comprehensive exploration of complex topics. Each agent contributes to the conversation by offering unique viewpoints, while the recorder agent maintains detailed logs of insights, consensus, and disputes. The summarization agent then compiles these discussions into a coherent summary, highlighting key points and unresolved issues. The experimental results demonstrate the effectiveness of this approach, and ME-RAG achieved better performance compared to the baseline models on various benchmarks. This method provides a systematic framework for leveraging multi-agent dialogue to improve information synthesis and decision-making in RAG systems.