Researchers at Stanford University have introduced Paper2Agent (P2A), a framework that transforms scientific papers into interactive AI agents. Led by computer scientist and biomedical data science professor James Zou, the project converts static research documents into dynamic entities that can engage in discussions, reproduce analyses, and apply findings to new datasets, enabling a paper to serve as an active “virtual author.” The process involves creating a Model Context Protocol (MCP) server incorporating a paper’s manuscript, supplementary materials, code, datasets, and analysis workflows, which then interacts with a large language model (LLM) agent to respond to natural-language inquiries, run demonstrations, and replicate analyses. Zou said the shift from traditional static formats to a more interactive platform could facilitate faster scientific discoveries and improve access to research. Concerns about accuracy, particularly AI hallucinations, remain, and the researchers emphasize rigorous evaluation, framing P2A as a tool to support scientific inquiry rather than a definitive source of truth, with each tool validated against original results to minimize errors and ensure reproducibility. Early testing converted 74 out of 100 computational biology papers into agents, with challenges including incomplete codebases and insufficient documentation. The open-source framework invites community collaboration, an online prototype is available, and researchers plan a platform for paper agents to collaborate on new scientific questions.