Paper2Agent turns research papers into runnable AI agents

Paper2Agent built 22 AlphaGenome tools in around 45 minutes for US $14, and the resulting AI agent answered 15 tutorial-derived queries with 98.7 ± 1.3% accuracy, against 82.7 ± 3.4% for Claude given direct access to the same code repository. The comparison sits at the heart of a paper by Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou, published in Nature on 16 September 2026.

"Here we introduce Paper2Agent, an automated framework that converts research papers into artificial intelligence (AI) agents," the abstract says. "Paper2Agent transforms research output from passive artefacts into active systems that accelerate use and discovery." The system reads a paper together with its codebase, constructs a Model Context Protocol (MCP) server that exposes the paper's methods as callable tools, then runs its own tests to harden them, so a chat agent can execute the workflow on demand instead of asking the reader to install the repo.

On a Scanpy agent, the same pipeline produced seven validated tools in around 45 minutes for US $13. Scaled up, the authors ran Paper2Agent against 100 computational biology papers: 74 were successfully agentified, yielding 599 proposed tools of which 593 passed automated validation, and the resulting agents reached 91.2 ± 1.6% accuracy across 300 tutorial-derived questions. Across 42 execution-based tasks from 10 non-biology computational papers, accuracy was 98.1 ± 0.8%. Median runtime fell by 1.9× and 3.1× relative to Claude + Repo and Biomni.

The framing the authors reach for is a role, not a format: Paper2Agent, they write, converts a paper into an agent that "functions as a virtual corresponding author." Five of the researchers we track posted the Nature link the same day.