Paper2Agent turns research papers into interactive AI agents

Paper2Agent, a framework published in Nature on September 16, converts research papers into AI agents that function as a "virtual corresponding author," letting a chat assistant like Claude run the paper's methods, fetch its datasets, and execute its workflows through natural-language queries.

Jiacheng Miao, James Zou, Jonathan K. Pritchard and colleagues tested the pipeline on 100 computational biology papers and successfully agentified 74, generating 599 tools of which 593 passed validation. The flagship AlphaGenome paper agent hit 98.7% accuracy on tutorial-derived queries and 100% on novel ones, with a 1.9× runtime reduction versus direct repository access.

The framing in the paper is that today "papers are fundamentally passive objects: a reader must discover the paper...parse its contributions and manually determine how to apply them to their own work." Paper2Agent, the authors write, "converts a paper into an AI agent that functions as a virtual corresponding author, exposing its manuscript, supplementary materials, datasets, code and workflows as active, agent-native knowledge rather than static text." The plumbing is the Model Context Protocol; the authors built agents around AlphaGenome (22 tools), Scanpy (7 validated tools), TISSUE, and datasets from Perturb-seq and MPRA-coupled scCRISPRi screens.

A substantial fraction of repositories still could not be agentified, "typically owing to incomplete codebases, missing documentation or unresolvable environment configurations." The authors add that maintenance is not a bug but a design feature: "Paper agents require ongoing maintenance as upstream codebases and dependencies evolve; we view this as an inherent feature of publishing executable research." The discussion also flags "security, intellectual property and attribution challenges that warrant careful handling," without resolving them.