Stanford's Paper2Agent turns papers into collaborating AI agents

Stanford Medicine researchers have built a system that turns scientific papers into AI agents that can talk to each other, and in one test, surface a link the original authors never made. Two agents built from unrelated studies, a genetic mutation prediction tool and a genome-wide association study of ADHD risk, identified a previously unreported connection: a molecular variant near the gene MPHOSPH9 associated with increased ADHD risk, according to Stanford Medicine.

The system is called Paper2Agent. It sends "worker agents" to reproduce a paper's research from scratch in a virtual environment, capturing methodology and know-how, then wraps the result (text, figures, data) as an interactive entity another agent can query. The team, led by postdoctoral scholar Jiacheng Miao and associate professor of biomedical data science James Zou, has built more than 100 such agents. The work was published in Nature on September 16, 2026.

"For essentially all of human history, the way that we represent knowledge is in the form of these very passive artifacts," Zou said. "Instead of having only passive artifacts, why don't we convert each static record into an active embodiment of knowledge?"

The plumbing runs on Model Context Protocol, which stores a paper's sections in separate folders so agents can find what they need. "An MCP lets AI essentially represent a paper PDF in a form that's easy for agents to access, almost like a filing system," Zou said.

Zou pictures millions of paper agents finding overlapping work at scale, pairs of studies whose authors would otherwise have to stumble into each other. He is blunt about the credit question: "It's still important to attribute the final discoveries and reference them back to original papers and original human authors."