Stanford's Paper2Agent turns 74 of 100 biology papers into agents

Paper2Agent, a framework from a Stanford-led team, converted 74 of 100 computational biology papers into working AI agents that answer natural-language questions by running the papers' own code, according to reporting in Nature News. The pipeline generated 599 candidate tools across those papers, and 593 passed automated validation.

The system, described by Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou, wraps a paper's manuscript, code and datasets behind a Model Context Protocol server that a chat client such as Claude Code can call. Building an agent for the AlphaGenome paper produced 22 validated tools in roughly 45 minutes at about US$14. That agent scored 98.7% on questions drawn from AlphaGenome's own tutorials and 100% on novel genomic queries, against 82.7% and 78.7% for Claude Code pointed at the raw repository.

'Knowledge should not be static records,' James Zou, a computer scientist at Stanford, told Nature News. 'It really should be dynamic and interactive.' Zou also framed the conversion itself as a quality check: 'Agentification itself is a useful certificate that says, "This work is relatively complete."'

The 26 papers that could not be converted failed because their code was missing, their dependencies broke, or their scripts were too narrowly written to generalize. Outside voices were qualified in their praise. Dongping Chen, a computer scientist at the University of Maryland, called the approach 'quite compelling.' Olivier Elemento of Weill Cornell Medicine described it as 'a real advance in terms of how we think about the publication process, with AI at the center.'

Zou proposed adding an 'agent availability' statement to papers, alongside the existing data and code statements. Three researchers on our Who's Who watchlist posted the piece the day it appeared.