Stanford's Paper2Agent turns research papers into callable AI agents

Paper2Agent, developed by James Zou's group at Stanford and reported in Nature on 16 September, ingests a paper alongside its code, datasets and supplementary materials and deposits the package on an MCP server so any language model can call the paper's methods as tools. Three of the researchers we track posted it the same week it landed.

Tested on the AlphaGenome paper, which predicts properties of DNA sequences, the group built a working agent in about 45 minutes at roughly US$14 in compute. It answered genetics questions with near-perfect accuracy and outperformed Biomni, a competing biomedical AI tool with access to dozens of databases. In one rerun on cholesterol-related genetic variants, the agent identified a different causal gene than the original AlphaGenome paper.

Zou, a computer scientist at Stanford, framed the wider ambition plainly: turning static papers into dynamic sources of information "can help us to reimagine what knowledge looks like in the future."