Paper Debugger: An Editor-Native Multi-Agent Framework for Agentic Academic Writing in Overleaf

Large Language Models (LLMs) have transformed academic writing, yet existing tools operate as external applications that disrupt cognitive flow through repetitive copy-paste cycles. This fragmentation prevents AI systems from understanding document structure, citation dependencies, and LaTeX-specific context. We present Paper Debugger, an editor-native multi-agent framework integrated directly into Overleaf via a Chrome extension. Leveraging Kubernetes orchestration and the Model Context Protocol (MCP), Paper Debugger deploys specialized agents for grammar refinement, structured critique, citation verification, and literature retrieval—while maintaining full revision transparency through deterministic diff-based patches. In a pilot study with 25 researchers, the system reduced manual formatting time by 34% (p < 0.05) and achieved a System Usability Scale (SUS) score of 78.4, indicating above-average user satisfaction. Unlike traditional external tools, Paper Debugger preserves workflow continuity and enables contextual, structure-aware AI assistance within the native authoring environment.