Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PaperDebugger: A Plugin-Based Multi-Agent System for In-Editor Academic Writing, Review, and Editing

Junyi Hou, Andre Lin Huikai, Nuo Chen, Yiwei Gong, Bingsheng He

TL;DR

PaperDebugger addresses the gap where AI writing assistance operates outside the editor by embedding a multi-agent, in-editor system into Overleaf. It combines a Chrome extension, Kubernetes backend, and MCP/XtraMCP tooling to support inline critique, literature retrieval, and diff-based edits with parallel agent execution and stateful revision provenance. The architecture enables end-to-end workflows—from micro-level edits to deep literature synthesis—directly in the writing environment, demonstrated through real-world deployment and analytics. The work provides evidence of feasibility, usability, and practical impact for editor-native, agentic academic writing assistance. Overall, it demonstrates a significant step toward integrated, context-aware AI collaboration in scholarly writing.

Abstract

Large language models are increasingly embedded into academic writing workflows, yet existing assistants remain external to the editor, preventing deep interaction with document state, structure, and revision history. This separation makes it impossible to support agentic, context-aware operations directly within LaTeX editors such as Overleaf. We present PaperDebugger, an in-editor, multi-agent, and plugin-based academic writing assistant that brings LLM-driven reasoning directly into the writing environment. Enabling such in-editor interaction is technically non-trivial: it requires reliable bidirectional synchronization with the editor, fine-grained version control and patching, secure state management, multi-agent scheduling, and extensible communication with external tools. PaperDebugger addresses these challenges through a Chrome-approved extension, a Kubernetes-native orchestration layer, and a Model Context Protocol (MCP) toolchain that integrates literature search, reference lookup, document scoring, and revision pipelines. Our demo showcases a fully integrated workflow, including localized edits, structured reviews, parallel agent execution, and diff-based updates, encapsulated within a minimal-intrusion user interface (UI). Early aggregated analytics demonstrate active user engagement and validate the practicality of an editor-native, agentic writing assistant. More details about this demo and video could be found at https://github.com/PaperDebugger/PaperDebugger.

PaperDebugger: A Plugin-Based Multi-Agent System for In-Editor Academic Writing, Review, and Editing

TL;DR

PaperDebugger addresses the gap where AI writing assistance operates outside the editor by embedding a multi-agent, in-editor system into Overleaf. It combines a Chrome extension, Kubernetes backend, and MCP/XtraMCP tooling to support inline critique, literature retrieval, and diff-based edits with parallel agent execution and stateful revision provenance. The architecture enables end-to-end workflows—from micro-level edits to deep literature synthesis—directly in the writing environment, demonstrated through real-world deployment and analytics. The work provides evidence of feasibility, usability, and practical impact for editor-native, agentic academic writing assistance. Overall, it demonstrates a significant step toward integrated, context-aware AI collaboration in scholarly writing.

Abstract

Large language models are increasingly embedded into academic writing workflows, yet existing assistants remain external to the editor, preventing deep interaction with document state, structure, and revision history. This separation makes it impossible to support agentic, context-aware operations directly within LaTeX editors such as Overleaf. We present PaperDebugger, an in-editor, multi-agent, and plugin-based academic writing assistant that brings LLM-driven reasoning directly into the writing environment. Enabling such in-editor interaction is technically non-trivial: it requires reliable bidirectional synchronization with the editor, fine-grained version control and patching, secure state management, multi-agent scheduling, and extensible communication with external tools. PaperDebugger addresses these challenges through a Chrome-approved extension, a Kubernetes-native orchestration layer, and a Model Context Protocol (MCP) toolchain that integrates literature search, reference lookup, document scoring, and revision pipelines. Our demo showcases a fully integrated workflow, including localized edits, structured reviews, parallel agent execution, and diff-based updates, encapsulated within a minimal-intrusion user interface (UI). Early aggregated analytics demonstrate active user engagement and validate the practicality of an editor-native, agentic writing assistant. More details about this demo and video could be found at https://github.com/PaperDebugger/PaperDebugger.

Paper Structure

This paper contains 11 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. System Overview
  3. Architecture Overview
  4. Agentic Design
  5. Usage Analytics
  6. Real-World Adoption
  7. Interaction Patterns
  8. Demonstration
  9. In-Editor Editing and Patch
  10. Deep Research and Comparative Analysis
  11. Conclusion

Figures (4)

  • Figure 1: Overall architecture of PaperDebugger, consisting of the presentation layer, backend layer, agent layer, protocol layer and infrastructure.
  • Figure 2: PaperDebugger end-to-end workflow: The extension captures user actions and sends them to the PaperDebugger server, which coordinates built-in agents or specialized agents on the XtraMCP server.
  • Figure 3: In-editor editing workflow in PaperDebugger. (1) Select a span of LaTeX in Overleaf. (2) Add the selection to the PaperDebugger panel. (3) Specify the critique request. (4) Trigger the agentic pipeline. (5) Review and apply the returned before--after patches.
  • Figure 4: Example of an end-to-end research-use scenario supported by PaperDebugger. The system integrates XtraMCP (a) deep research, (b) related-paper retrieval, and (c) section enhancer to help authors understand, compare, and refine academic content within the editor.