Paper2Agent turns 74 of 100 bio papers into AI agents

Paper2Agent, an automated framework published in Nature, turned 74 of 100 computational biology papers into working AI agents without manual intervention.

The system reads a paper and its codebase, then builds a Model Context Protocol server that exposes the paper's tools and workflows to chat agents like Claude Code. Users query in natural language instead of installing dependencies and configuring environments. For AlphaGenome, Paper2Agent generated 22 tools in about 45 minutes at a cost of $14; for Scanpy, 7 tools in the same window for $13.

"Paper2Agent transforms research output from passive artefacts into active systems that accelerate use and discovery," the authors write. Papers, they add, "become agent-native research objects rather than static documents."

On accuracy, the AlphaGenome agent scored 98.7 ± 1.3% on tutorial-derived queries and 100.0 ± 0.0% on novel queries, against Claude + Repo at 82.7 ± 3.4% and 78.7 ± 4.4%, and Biomni at 37.3 ± 4.0% and 56.0 ± 3.4%. In the larger sweep, 593 of 599 proposed tools passed validation across 100 computational biology papers. The framework ran 1.9x faster than Claude + Repo and 3.1x faster than Biomni, at $0.20 per query versus $0.38. On 10 non-biology papers spanning tools like SAM2, TabPFN and CausalImpact, accuracy was 98.1 ± 0.8% across 42 execution tasks.

In one demonstration, multiple paper-agents collaborated to prioritize causal genes for psoriasis at the rs887314 locus, flagging GPR137 with a Spearman correlation of 0.613 (p=3.79x10⁻³). Four researchers we track shared the paper on the day it appeared.