Stanford's Paper2Agent, Published in Nature, Turns Papers into Interactive AI Agents

Stanford's James Zou team (first author Jiacheng Miao and others) published Paper2Agent in Nature, a framework that converts research papers from static texts into interactive, reusable AI agents acting as 'virtual corresponding authors.' James Zou detailed the technical motivation in a thread, while Eric Topol and Valerio Capraro also highlighted the work, alongside a Nature news report.

Confirmed

• Paper2Agent is published in Nature. It uses multiple agents to analyze a paper and its codebase (manuscript, code, data, supplementary materials), build a corresponding MCP server, and verify the method by generating and running tests.
• James Zou explained the motivation: papers are written for humans, and feeding a PDF directly to an LLM loses key insights. Instead, the AI builds a 'virtual lab,' running code, testing methods, and reproducing results to gain hands-on experience rather than only reading text.
• The output is an auto-generated 'paper MCP' encapsulating the paper's know-how, tools, and data. Zou reports that giving an LLM this paper MCP yields significantly higher accuracy and efficiency than providing the original paper plus code repository; the MCP server can be accessed by any AI agent.
• Capabilities: instant Q&A about a paper to help researchers enter unfamiliar fields; applying a paper's methods to new data; multiple paper agents collaborating and making new discoveries. Valerio Capraro demonstrated combining agents derived from different papers and noted researchers can interrogate methods like contacting a corresponding author.
• @bravoabad added quantitative data: in one example, Paper2Agent generated 22 tools in about 45 minutes, achieving 100% accuracy on their benchmark.

Why it matters

• Scientific knowledge has traditionally been stored as passive papers; Paper2Agent turns papers into conversational, executable, composable agents, transforming how knowledge is stored and reused. It dramatically lowers the barrier to cross-domain method transfer.
• By running code to reproduce results, the 'virtual author' is not just a Q&A tool but can participate in real research workflows. Nature's news coverage noted such tools turn static literature into interactive knowledge objects, potentially reshaping how researchers digest the literature.

2026-09-16 ~ 2026-09-18 · 18 related posts

• Paper2Agent turns research papers into interactive AI agents (Nature) — EricTopol · 2026-09-16
• [source] Paper2Agent detail: AI builds a virtual lab to reproduce papers — james_y_zou · 2026-09-17
• Paper2Agent's auto-generated paper MCP beats paper + code repo — james_y_zou · 2026-09-17
• Stanford's James Zou Shows Paper MCPs: Agents From Different Papers Collaborating on Research — james_y_zou · 2026-09-17
• Paper2Agent turns research papers into interactive AI agents, Nature paper shows — ValerioCapraro · 2026-09-17
• Paper2Agent in Nature converts research papers into MCP-based AI agents automatically — ValerioCapraro · 2026-09-17
• AI tool turns any research paper into an 'agent' that answers complex queries — Fcking_Chuck · 2026-09-17
• [source] Paper2Agent turns papers into executable MCP agents: 22 validated tools in 45 minutes, 100% on novel benchmarks — bravo_abad · 2026-09-17
• Stanford's Paper2Agent turns scientific papers into collaborating AI agents — burny_tech · 2026-09-18

9 near-duplicate retellings: james_y_zou · james_y_zou · EricTopol · james_y_zou · james_y_zou · james_y_zou · soumitrashukla9 · krishnan · james_y_zou