Stanford's Paper2Agent turns research papers into callable AI agents

Stanford researchers have built a tool that reads a scientific paper's text, code and data and turns it into a working AI agent other scientists can query in plain language. Nature reported on 16 September 2026 that the tool, Paper2Agent, deposits a paper's contents onto an MCP server and then lets a team of AI agents autonomously build tools that apply the paper's methods, which any large language model can then call.

Tested on the AlphaGenome paper, which predicts properties of DNA sequences, the group built a working agent in about 45 minutes with computing power costing US$14, and it answered genetics questions with near-perfect accuracy. The AlphaGenome agent outscored Biomni, a tool developed by academic researchers that draws on dozens of databases.

The demonstration drawing attention is a cholesterol case. Asked to identify a gene explaining why a DNA change is linked to bad cholesterol, the agent named a different causal gene from the one pinpointed in the original AlphaGenome paper. "AlphaGenome's data on genetic variants support both hypotheses," said James Zou, the Stanford researcher who led the work, arguing scientists can use the tool to re-evaluate published conclusions without designing entirely new experiments.

The team also let two paper-agents collaborate, one built from a genome-mutation prediction paper and one from an ADHD genome-wide association study, and the pair flagged a molecular variant near the MPHOSPH9 gene associated with increased ADHD risk that had not previously been documented in the scientific literature. Zou says the approach "can help us to reimagine what knowledge looks like in the future." Three researchers we follow shared the paper the day it landed.