With the rapid growth of the web novel market, the increasing difficulty of creating long-form narratives has exacerbated writers' cognitive load due to setting errors. To address this issue, this study proposes Co-Narrator, a multi-agent system based...
With the rapid growth of the web novel market, the increasing difficulty of creating long-form narratives has exacerbated writers' cognitive load due to setting errors. To address this issue, this study proposes Co-Narrator, a multi-agent system based on Large Language Models (LLMs). The proposed system integrates Retrieval-Augmented Generation (RAG) and the Model Context Protocol (MCP) to effectively identify narrative consistency issues and logical errors within extensive texts. Specifically, a precision-oriented detection strategy was adopted to minimize writer fatigue. Experimental results on official benchmarks and web novel datasets show that the proposed model improves Precision from 0.42 to 0.65 (+0.23), indicating its superiority in long-context processing. This study redefines agents—which utilize various tools tailored to user needs—not as replacements for creativity but as active assistants, suggesting a new approach for maintaining quality in serialized narratives.