ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
TopicDisconfirmAI EditorialUpdated 2026-11-22 19:41

Paper2Agent Framework

3 steps1 pieces4 Sources

The Paper2Agent framework, developed by researchers at Stanford University including Jiacheng Miao and James Zou, converts static research papers into interactive AI agents. Published in Nature on September 16, 2026, the system packages a paper’s codebase, data, and methods as a Model Context Protocol (MCP) server, allowing users to query and execute scientific workflows through natural language rather than manual code adaptation.

The development timeline begins with the initial arXiv preprint submitted on September 8, 2025, which introduced the core concept of "agentifying" papers. This was followed by the peer-reviewed publication in Nature in September 2026, where the team demonstrated that 74 out of 100 computational biology papers could be successfully converted into working agents. These agents achieved 91.2% accuracy on benchmark questions, surpassing the 80.3% score of Claude Code directly accessing the raw repositories. Case studies included AlphaGenome for genomic variant interpretation and Scanpy for single-cell analysis.

This matters because it addresses the long-standing barrier of reproducibility in computational science. Traditionally, applying a new method requires locating repositories, installing dependencies, and configuring environments—a process described as "passive and fragmented." Paper2Agent proposes that if a paper can be successfully converted into an executable agent, its robustness is validated. The disagreement lies in scalability: while proponents argue this creates a "virtual corresponding author," critics note it is not guaranteed to work on all papers, particularly those with complex or undocumented dependencies.

To watch next, look for the adoption of "agent availability" sections in academic journals alongside traditional code availability statements. Additionally, monitor whether major platforms like Hugging Face or AWS see increased hosting of these MCP servers, and track citations of the Nature paper to gauge industry uptake.

How it got here

  1. 2025-09-08

    September 8, 2025: The Paper2Agent framework is first introduced via an arXiv preprint by Jiacheng Miao et al., outlining the automated conversion of papers into MCP servers.

Sign in / Sign up
ReadingResearchRadarInvestment framework
arxiv.org ↗
  • 2026-09-16

    September 16, 2026: The peer-reviewed article 'Reimagining research papers as interactive and reliable AI agents' is published in Nature, detailing case studies on AlphaGenome and Scanpy.

    nature.com ↗
  • 2026-11-19

    November 19, 2026: Snorkel AI hosts a reading group session featuring Jiacheng Miao, presenting performance metrics showing 91.2% accuracy for paper agents versus 80.3% for direct code access.

    snorkel.ai ↗
  • What is still unsettled

    • How will academic publishers integrate 'agent availability' into standard submission guidelines?
    • Can the framework reliably handle papers with highly proprietary or non-open-source dependencies?
    • What are the long-term maintenance costs for keeping hosted paper agents updated as underlying libraries evolve?

    Sources

    • snorkel.aiSnorkel AI Blog↗
    • nature.comNature Article↗
    • arxiv.orgarXiv Preprint↗
    • medium.comMedium Analysis↗

    Coverage in this topic

    • Stanford HAI Postdoctoral Fellow Jiacheng Miao Proposes: Whether a Paper Can Become an Executable Agent Is Itself a Reproducibility Metric2026-11-20News

      Jiacheng Miao (Stanford Institute for Human-Centered AI) introduced the Paper2Agent framework at a Snorkel AI Reading Group session: packaging papers and their codebases as MCP servers, enabling agents to learn to execute paper methods through a build-and-test loop. The team converted 100 computational biology papers into 74 usable agents, achieving a score of 91.2% on 300 benchmark questions, compared to 80.3% for Claude Code directly accessing the code repositories (both using Sonnet 4). Miao’s core argument is that whether a paper can be successfully executed by an agent serves as a new metric for reproducibility.