Nature paper: Paper2Agent turns research articles into AI agents

Of 100 computational biology papers put through the automated pipeline described in Nature, 74 came out the other side as working AI agents, together generating 599 tools of which 593 passed automated validation.

"Here we introduce Paper2Agent, an automated framework that converts research papers into artificial intelligence (AI) agents," the authors write. The system, they say, transforms "research output from passive artefacts into active systems that accelerate use and discovery."

Each generated agent runs on a Model Context Protocol server and behaves, in the paper's phrasing, as "an interactive expert on the corresponding paper, capable of demonstrating, applying and adapting its methods to new projects." The team, led by Jiacheng Miao with Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou, ran three case studies: AlphaGenome for genomic interpretation, Scanpy for single-cell analysis, and TISSUE for spatial transcriptomics.

The numbers are cheap. The AlphaGenome agent generated 22 MCP tools in 45 minutes at a compute cost of $14. Scanpy produced 7 tools on a similar budget. On benchmark questions the resulting agents hit 91.2 ± 1.6% accuracy, with 98.7 ± 1.3% on tutorial queries and 100.0 ± 0.0% on novel ones, running 1.9× faster than a Claude-plus-repository baseline.

The paper also reports that "paper agents can interact directly with each other," which the authors frame as autonomous knowledge synthesis across the literature. Five researchers on our Who's Who tracker have shared the link.

The write-up carries headline accuracy for the biology corpus but no equivalent breakout in its summary for the 26 data-focused papers or the 10 non-biology computational papers Paper2Agent was also run against.