Paper2Agent turns 74 of 100 biology papers into AI agents

Paper2Agent, described in Nature this month, converts a scientific paper into an AI agent that acts as a 'virtual corresponding author,' letting a user interrogate the manuscript's methods, run its code, and query its datasets by chat. The authors tested it across 100 computational biology papers and got working agents out for 74.

Inside those 74 papers, 593 of 599 proposed tools passed validation. On tutorial-derived benchmarks the agents hit 91.2% accuracy on average, and 98.7% on queries drawn from AlphaGenome tutorials. The system ran between 1.9x and 3.1x faster than the papers' baselines.

The five authors, Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou, pitch the shift plainly. 'Paper2Agent addresses this challenge by converting a paper into an AI agent that functions as a virtual corresponding author,' they write, before going further: 'Just as many journals now require data and code availability sections, we anticipate the emergence of an "agent availability" section.'

The pipeline runs on Model Context Protocol (MCP) with Claude Code and Claude Sonnet 4, and packages agents for Hugging Face Spaces. That anchors the framework to a specific vendor stack, which is a real dependency for any lab considering it for its own paper.