Stanford researchers have developed Paper2Agent, a framework that converts scientific papers and their associated code, data and workflows into interactive AI agents. The agents can explain findings, reproduce analyses, apply research methods to new datasets and collaborate with other paper-based agents, potentially making scientific work easier to access, verify and reuse.

Paper2Agent creates a Model Context Protocol server exposing a study’s research tools and resources for an AI system to operate through natural-language requests. In tests across 136 papers, including 100 in computational biology, 74 were successfully converted; failures were mainly linked to incomplete code, missing documentation or difficult software environments. The system is open source, but researchers caution that its outputs still require checking, particularly because sensitive data may be sent to the connected AI model and hallucinations remain possible.