Paper2Agent: Stanford Framework That Turns Research Papers into AI Agents

Forum topic · ✨步子哥 · 2025-10-07

Summary

Paper2Agent is an automated framework proposed by Stanford University researchers that converts scientific papers into interactive 'research assistant' AI agents, moving science from passive reading toward conversational use of published methods. Built on the MCP (Model Context Protocol) ecosystem, the framework consists of two key components: Paper2MCP and an agent layer. The MCP server packages a paper's contributions into three elements—executable MCP tools encapsulating methodological innovations, MCP resources holding the paper text, code repository, and supplementary materials, and MCP prompts guiding multi-step scientific workflows. The conversion pipeline identifies key contributions (datasets, methods, models, or workflows), wraps them as an MCP server with explicit inputs, outputs, and usage instructions, then connects it to an LLM agent for natural language queries and autonomous execution. Paper agents are interactive, requiring no programming skills, and reliable, since tools are validated against example datasets with pinned configurations for reproducibility, and support multi-agent collaboration. A featured case study is the AlphaGenome agent.

Paper2Agent: From Reading Papers to Conversing with Them

Paper2Agent is an automated framework proposed by Stanford University researchers that turns scientific papers into "research assistant" AI agents, allowing published academic work to be directly "invoked" and shifting research from *reading* papers to *conversing* with them. The framework supports multi-agent collaboration.

Key Characteristics of Paper Agents

Interactive: users can run complex scientific analyses via natural language, with no programming required
Reliable and reproducible: tools are validated against example datasets, and configurations are pinned to ensure reproducibility

Technical Implementation

The framework is built on the MCP (Model Context Protocol) ecosystem, with two key components:

1. Paper2MCP — converts a paper into an MCP server 2. Agent layer — LLM agents that consume the MCP server

Core Components of the MCP Server

| Component | Role | |---|---| | MCP Tools | Executable functions encapsulating the paper's methodological innovations | | MCP Resources | Static repositories: paper text, code repository, and supplementary materials | | MCP Prompts | Refined instruction templates that guide agents through multi-step scientific workflows |

Conversion Pipeline

1. Identify key contributions — datasets, methods, models, or workflows 2. Wrap as an MCP server — with explicit inputs, outputs, and usage instructions 3. Connect to an LLM agent — enabling natural language queries and autonomous execution

Case Study: AlphaGenome Agent

The deck includes a case study on an AlphaGenome agent, demonstrating how a paper's computational methods can be exposed as an interactive agent that researchers query in natural language while retaining validated, reproducible behavior.

Original post images (slide deck assets):