What Is Paper2Agent?
Paper2Agent is an AI system developed by Stanford Medicine researchers that transforms scientific manuscripts into interactive AI agents.
Instead of treating a research paper as a static PDF, Paper2Agent uses the paper's text, figures, associated code and data to create an agent capable of interacting with readers and other agents. Stanford says the system can answer questions, apply methods from a paper to new data and participate in agent to agent conversations. (med.stanford.edu)
The research was led by Jiacheng Miao and James Zou, and the work was published in Nature on September 16, 2026.
How Static Papers Become AI Agents
Paper2Agent uses multiple AI worker agents to examine a research paper and its associated resources.
The system does not simply summarize the PDF. The agents attempt to reproduce the research described by the manuscript, using that process to capture practical information about methods, dependencies, workflows and analysis procedures.
This information is organized through the Model Context Protocol, or MCP.
Stanford describes MCP as a way of representing the contents and functionality of a paper in a structure that AI systems can access more directly.
What Can a Paper Agent Do?
A resulting paper agent can perform several tasks:
• Answer questions about the research
• Explain methods
• Apply documented methods to new data
• Interact with associated tools
• Communicate with other paper agents
• Help reproduce computational results
The distinction is important. A paper agent is intended to expose usable research knowledge rather than simply generate a conversational summary.
Note: Paper2Agent does not eliminate the need for researchers. Stanford notes that manuscripts often omit failed experiments, undocumented decisions and other human judgment involved in research. Human authors can provide that missing context through interaction with the agent. (nature.com)
AI Agents Talking to Other Papers
One of the most interesting demonstrations involved two unrelated research papers.
Stanford converted both papers into agents and allowed them to interact. One agent contained knowledge about predicting genetic mutation effects, while another represented research involving ADHD risk.
The agents identified a potential connection involving a molecular variant near MPHOSPH9.
Stanford presents this as a demonstration of how paper agents could surface relationships that might otherwise require researchers from separate fields to discover one another manually.
How Researchers Can Access Paper2Agent
Paper2Agent is publicly available.
Researchers and developers can inspect the source code, follow the setup instructions and use the system to agentify research papers. The project also provides demonstrations and connectable MCP servers for selected research tools. (github.com)
The project therefore does not require researchers to wait for a closed academic platform before experimenting with the concept.
Paper2Agent Access Model
| Component | Availability |
|---|---|
| Paper2Agent source code | Public |
| Research methodology | Published |
| Paper agents | Demonstrations available |
| MCP servers | Selected examples available |
| Human researcher input | Supported |
| AI to AI collaboration | Demonstrated |
Why Interactive Manuscripts Matter
Traditional scientific papers are designed primarily for human reading.
Paper agents introduce another possibility: the paper becomes an active interface that can explain its methodology, execute supported analyses and connect its knowledge with other research.
This could eventually change how researchers search for methods and understand previous work. Instead of asking a conventional search engine to find relevant papers, future systems could allow specialized research agents to interact directly with the methods and data described in those papers.
Stanford says its team has already created more than 100 paper agents, with a longer term goal of scaling the concept to a much larger research network.
Final Take
Paper2Agent represents a shift from documents as static records to documents as interactive research interfaces.
Its importance is not simply that an AI can summarize a paper. The more ambitious idea is that a paper can become an agent capable of explaining its methods, reproducing analyses and interacting with other scientific knowledge.
With the project publicly available through GitHub and demonstrations already accessible, researchers can experiment with the system rather than treating it as a purely theoretical research concept.