methods-mcp

Structured methods extraction for academic papers, designed for the Worldwide AI Science Fellowship build challenge. Exposes eight tools that resolve paper URLs (arXiv, DOI) into canonical metadata, full text with section splits, Pydantic-validated methods objects (steps, reagents, equipment, analyses), and associated code repositories discovered via Papers With Code. The assess_repo_reproducibility tool scores GitHub repos without cloning by analyzing README presence, dependency files, notebooks, figure scripts, and maintenance signals through the GitHub REST API. The composite methods_repro_review tool runs the full pipeline in one call and returns a plain-English narrative alongside structured data. Uses Claude Sonnet 4 for extraction via tool-use validation with one repair attempt on schema failures. Reach for this when you need on-demand reproducibility verdicts that return in seconds rather than the 30-minute runtimes of execution-based pipelines.