Stanford team turns research papers into tested AI agents
A Stanford team led by Jiacheng Miao and James Zou has introduced Paper2Agent, a system that converts a research paper and its codebase into a Model Context Protocol (MCP) server. Published in Nature on 16 September 2026, the project lets compatible agents such as Claude Code run documented methods through natural-language requests. The MIT-licensed software can be installed for Claude Code or Codex, while prebuilt AlphaGenome, Scanpy and TISSUE servers are available on Hugging Face Spaces and through paper2agent.ai.
Paper2Agent creates isolated environments, scans tutorials, executes them, extracts parameterized tools and verifies their outputs. Validation requires expected files and numerical results to match within 3%; figures are compared using perceptual hashes. Failed tools are removed from the resulting server. In tests covering 300 questions from 74 papers, the system achieved 91.2% accuracy, compared with 80.3% for Sonnet 4 and 86.3% for Sonnet 4.6 paired with repository access. It converted 74 of 100 bioRxiv computational-biology papers, validating 593 of 599 proposed tools. An AlphaGenome server produced 22 validated tools in about 45 minutes at a cost of US $14.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.