Paper2Agent turns 74 of 100 biology papers into AI agents

Paper2Agent, a framework introduced by Jiacheng Miao, Jonathan K. Pritchard, James Zou and colleagues, converts scientific publications into Model Context Protocol servers that a chatbot can query in natural language. Applied to 100 computational biology papers, the Nature paper reports that 74 could be agentified, producing 599 proposed tools of which 593 passed automated validation. Across 300 benchmark questions the ensemble scored 91.2%.

On the AlphaGenome codebase specifically, the pipeline produced 22 validated tools in about 45 minutes for less than $15 in computing costs. The authors describe the result as "interactive experts on the corresponding paper, capable of demonstrating, applying and adapting its methods."

"Knowledge should not be static records. It really should be dynamic and interactive," Zou, a Stanford computer scientist, told IEEE Spectrum. He framed the pipeline as doubling as a filter on the underlying work: "Agentification itself is a useful certificate that says, 'This work is relatively complete and well documented.'"

Outside comment ran cautious rather than skeptical. Dongping Chen at the University of Maryland told the same outlet that "the idea of making papers more dynamic and executable through an agentic interface is quite compelling." Five of the researchers we track posted the paper's link within days of publication.

The paper does not say what disqualified the 26 papers that failed conversion.