Paper2Agent turns research papers into queryable AI agents

A team led by James Zou, Jonathan K. Pritchard and Jiacheng Miao has published a system in Nature that turns a scientific paper into an interactive AI agent in about 45 minutes at a cost of $14, running on a personal laptop.

The framework, called Paper2Agent, uses the Model Context Protocol to expose a paper's code, data and workflows as callable tools that a chat agent such as Claude Code can invoke through natural language. In the AlphaGenome case study, the pipeline "generated 22 AlphaGenome MCP tools, all of which passed automated validation, in around 45 min, costing US $14 on a personal laptop without human intervention."

On that case study the paper reports "98.7 ± 1.3% accuracy on 15 tutorial-derived queries and 100.0 ± 0.0% accuracy on 15 novel queries." Scaled across 100 computational biology papers, tutorial-derived accuracy came in at 91.2 ± 1.6%.

The authors also chain paper-agents together. In a psoriasis workflow, collaborating agents flagged GPR137 as a candidate causal gene, reporting that "Only GPR137 knockdown showed significant concordance with the CRE perturbation signature under stimulated conditions (Spearman correlation = 0.613, P = 3.79 × 10−3 at Stim8hr)."

The framing is explicit. The authors argue an agent-native paper "functions as a virtual corresponding author, exposing its manuscript, supplementary materials, datasets, code and workflows as active, agent-native knowledge rather than static text." Four AI researchers we track posted the paper's link.

The abstract never itemises where errors on the 100-paper benchmark came from, and the novel-query result is 15 items on a single paper. "Without human intervention" refers to the tool-generation loop, not the queries themselves.