Stanford University researchers Jiacheng Miao and James Zou have developed Paper2Agent, an open-source tool that transforms computational research papers and codebases into Model Context Protocol (MCP) servers. Published in Nature, the framework allows MCP-compatible agents like Claude Code to execute paper methods using natural language queries, acting as an automated corresponding author.
Paper2Agent uses an orchestrator running on Claude Code’s agent SDK to dispatch specialized sub-agents across a six-step generation pipeline. To ensure execution accuracy, the system enforces strict validation gates requiring output numerical values to match reference data within 3% and generated figures to meet perceptual hash thresholds. In benchmarking against tools like Biomni, Paper2Agent accelerated tutorial task execution times by up to 3.1x while incurring minimal API costs per server build.
The system is released under an MIT license, featuring prebuilt servers on Hugging Face Spaces and an online hosted portal. During multi-agent evaluations combining genomic and transcriptomic servers, Paper2Agent successfully generated and validated cross-dataset biological hypotheses, illustrating the potential for automated scientific discovery platforms.
Why it matters
Standardizes research paper execution by packaging complex academic codebases into production-ready Model Context Protocol (MCP) servers.
Reduces technical friction for R&D teams integrating third-party academic algorithms into existing operational pipelines.
Provides a reproducible blueprint for automated multi-agent code translation, test-driven validation, and tool creation.
Source: marktechpost.com
