Paper2Agent: Stanford Reimagining Research Papers as Interactive AI Agents (arxiv.org)

Stanford’s Paper2Agent is an automated framework that turns research papers and their code into interactive AI agents, effectively converting static publications into “knowledgeable research assistants.” The system uses multiple specialized agents to analyze a paper and its codebase, build a Model Context Protocol (MCP) server that encapsulates the paper’s models, data access and workflows, and then iteratively generates and runs tests to refine and harden the MCP. These MCPs can be hooked up to chat-driven tools (the authors demonstrate integration with Claude Code) so users can ask complex scientific questions in natural language and have the agent invoke the original paper’s tools and pipelines to answer them. Significance for AI/ML: Paper2Agent lowers barriers to adoption, reproducibility and reuse by automating the tedious work of understanding, adapting, and running others’ methods. In detailed case studies the framework produced agents that reproduced original results and handled novel queries—examples include an AlphaGenome agent for genomic-variant interpretation and ScanPy/TISSUE-based agents for single-cell and spatial transcriptomics. Technically, the contribution lies in formalizing a protocol (MCP) for packaging paper functionality, using multi-agent analysis plus test-driven refinement to improve reliability, and enabling natural-language orchestration of research workflows—paving the way for more interactive, discoverable, and collaborative “AI co-scientists.”