Editorial illustration for Stanford Paper2Agent Converts Research to AI Agents via MCP Tools
Paper2Agent Turns Research Code Into AI Agents
Stanford Paper2Agent Converts Research to AI Agents via MCP Tools
Most computational biology papers come with a GitHub repository nobody outside the lab ever gets running. The code demands a specific Python version, a dozen dependencies that conflict with whatever's already installed, and enough undocumented assumptions that reproducing even the paper's own figures can take days. That friction means genuinely useful methods sit in PDFs, cited but not used.
Jiacheng Miao and James Zou, working out of Stanford, built a system to close that gap. Paper2Agent, published in Nature on September 16, 2026, takes a paper and its codebase and converts them into a Model Context Protocol server, the same protocol that lets tools like Claude Code talk to external software. Once converted, anyone can query the paper's methods in plain language and get results back, no cloning or environment setup required.
The team calls it a virtual corresponding author, standing in for the person you'd normally have to email with questions about getting the code to run. It's already deployable: the project is MIT-licensed, installs as a Claude Code or Codex skill, and has prebuilt servers for tools like AlphaGenome and Scanpy live on Hugging Face Spaces. A hosted version sits at paper2agent.ai. Getting there required a six-step pipeline with a validation process built to catch tools that don't actually work.
A Stanford team led by Jiacheng Miao and James Zou proposes a fix. Paper2Agent was published in Nature on 16 September 2026. It converts a paper and its codebase into a Model Context Protocol (MCP) server.
Why this matters
Paper2Agent points at a real bottleneck: methods papers that die in a GitHub repo nobody can install. Wrapping code as MCP tools, with resources for data and prompts for workflow order, means the agent knows not just what a function does but when to call it. That's a meaningful distinction for anyone who's tried to reproduce a bioinformatics pipeline from a methods section alone.
For researchers, this could mean actually reusing someone else's pipeline instead of re-implementing it badly. For developers, it's a concrete pattern for turning any codebase into an agent-callable service rather than a black box you have to reverse-engineer. For founders building on top of research, it lowers the cost of prototyping on published methods before committing engineering time.
Worth watching: this relied on Claude Sonnet 4 specifically, so how well it generalizes across models and messier, less-documented codebases than the ones Stanford picked is still an open question. Reproducibility claims need testing on papers nobody curated for a demo.
Related Reading
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Anthropic and OpenAI Propose Safety Evaluator Access to Staff
From Archives TensorRT Edge-LLM Runs MLPerf Edge Agentic Benchmark 6.4x Faster on Jetson AGX Thor
Common Questions Answered
What problem does Paper2Agent solve for computational biology research?
Paper2Agent addresses the issue of research methods sitting unused in GitHub repositories due to complex dependencies, undocumented assumptions, and difficult setup requirements that prevent researchers outside the original lab from reproducing or using the code. By converting papers and their codebases into accessible AI agents, it enables genuine methods to be utilized rather than merely cited in academic literature.
How does Paper2Agent convert research papers into AI agents?
Paper2Agent converts a paper and its codebase into a Model Context Protocol (MCP) server, which wraps the code as MCP tools with resources for data and prompts for workflow order. This approach allows the AI agent to understand not just what each function does, but also when and how to call it in the proper sequence.
Who developed Paper2Agent and where was it published?
Paper2Agent was developed by Jiacheng Miao and James Zou at Stanford University and was published in Nature on September 16, 2026. The system represents a collaborative effort to bridge the gap between published research methods and their practical implementation.
What is the key advantage of using MCP tools for wrapping research code?
The key advantage of using MCP tools is that they provide contextual understanding beyond just function documentation—the agent knows not only what each function does but also the proper workflow order and when to call each function. This distinction is particularly meaningful for complex bioinformatics pipelines where sequence and dependencies are critical to successful reproduction.
Further Reading
- Paper2Agent: Reimagining Research Papers As Interactive and Reliable AI Agents - arXiv
- Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data - MarkTechPost
- Paper2Agent Converts Scientific Papers into Interactive AI Agents - InfoQ
- AI tool turns any paper into an agent that can collaborate and answer complex queries - Nature
- LiveMCPBench: Can Agents Navigate an Ocean of MCP Tools - AlphaXiv
