Paper2Agent
Open-source Stanford project that turns a research paper and its code into a tested MCP server that AI agents can call.
What it does
Paper2Agent is an open-source research project from Stanford University, described in an article in Nature (September 2026). It takes a scientific paper together with its code repository and builds an MCP server with tested tools and a skill from it. The conversion runs as a skill inside a coding agent such as Claude Code or Codex: parallel sub-agents analyse the code, set up an isolated environment and check the tools with tests. The result is a ZIP package with instructions that you connect to Claude Code, Codex or Gemini CLI and then use in natural language. The site paper2agent.ai, linked from the repository, also offers an app interface that works with a locally running Paper2Agent Runtime. Example servers for AlphaGenome, Scanpy and TISSUE run on Hugging Face.
Who it suits
Research teams, labs and data science departments, for example in biotech and pharma, that want to apply methods from publications to their own data. You should be comfortable with the command line, Python and Git.
Data protection for businesses
The conversion runs in your own environment. The paper, code and the content the agent reads are sent to the model provider behind your coding agent, whose rules on server location, retention and training apply. The project does not offer a data processing agreement, EU data hosting or business plans. The vendor does not name a privacy policy for paper2agent.ai or privacy information for the hosted example servers.
Limits
Tools are built only from existing repository code; without usable code there is no agent. The method has been tested mainly in computational biology. You need to check the agents’ results and hypotheses with domain expertise. Depending on the repository, you need R, GPUs or API keys. The project does not name guaranteed support; questions go through GitHub issues.