Paper2Agent turns Nature papers into callable AI agents

A Nature paper published 16 September 2026 introduces Paper2Agent, an automated framework that converts a research paper into an AI agent by wrapping its manuscript, code, data and workflows behind a Model Context Protocol server that a chat agent such as Claude Code can call. The authors — Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard, James Zou and colleagues — argue that under this treatment "papers become agent-native research objects rather than static documents."

Run over 100 computational biology papers, the pipeline produced working agents from 74 of them, generating 599 proposed tools of which 593 passed validation. On a benchmark drawn from AlphaGenome's own tutorials, the Paper2Agent-derived AlphaGenome agent hit 98.7% accuracy on tutorial queries and 100% on novel queries, against 82.7% for Claude Code pointed at the raw repository and 37.3% for Biomni. Median per-query cost fell to $0.20 from $0.38, at roughly 1.9× the speed.

To show what composition looks like, three paper-derived agents were chained together to prioritize a causal gene for psoriasis. They converged on GPR137 as the gene mediating the effect of the variant rs887314, a prediction the authors report was validated against independent experimental data.

The paper's forward claim is procedural as much as technical. "Just as many journals now require data and code availability sections, we anticipate the emergence of an 'agent availability' section," the authors write, framing the work as a proposal for how scientific communication should evolve. Five researchers in our Who's Who tracker posted the link the day it appeared.