Paper2Agent framework, validation and scale

The study introduces Paper2Agent, an automated framework that converts research papers into AI agents. It turns a paper’s manuscript, supplementary materials, datasets, code and workflows into active, agent-native knowledge rather than static text. The system analyses a paper and its codebase using multiple agents to construct a Model Context Protocol server, then generates and runs tests to refine MCP robustness. The resulting paper MCPs can connect to a chat agent to carry out scientific queries through natural language while invoking tools and workflows from the paper. Case studies include agents based on AlphaGenome, Scanpy and TISSUE. The authors report that these agents reproduce results from the original papers and answer novel user queries. Paper2Agent also created multiple collaborating agents that prioritized a causal gene for psoriasis. The authors describe this as a framework for knowledge dissemination and a collaborative ecosystem of AI co-scientists.

Paper2Agent is a multi-agent AI system that transforms research papers into interactive AI agents with minimal human input. The generated agents are designed to be interactive and easy to use, allowing complex analyses through natural-language prompts, and reliable and reproducible because each tool is validated against reference-code results and figures using example datasets and then locked. Each tool includes a code reference for transparency. The framework has Paper2MCP, which extracts information from papers and codebases to build remote MCP servers, and an agent layer that wraps each server as a paper-specific context provider. MCP tools encapsulate methodological contributions, resources hold manuscripts, code, supplementary materials and datasets, and prompts guide multi-step workflows. The workflow locates and downloads a codebase, sets up an isolated environment, discovers and executes tutorials, extracts reusable tools, tests and refines them, and assembles a deployable MCP server. Tools repeatedly failing validation are excluded. The resulting servers can be hosted remotely and connected to compatible language models.

The authors evaluated Paper2Agent on three corpora: 100 computational-biology papers, 26 data- and discovery-focused papers, and 10 non-biology computational papers spanning AI, statistics, econometrics, game theory and astrophysics. Among the 100 computational-biology papers, 74 were successfully agentified, yielding 599 proposed tools, of which 593 passed automated validation. Failure modes included missing executable code, missing data or model artefacts, environment or dependency failures and non-generalizable scripts. On 300 tutorial-derived benchmark questions, Paper2Agent with Sonnet 4 achieved 91.2 ± 1.6% accuracy, compared with 80.3 ± 2.3% for Claude Code with direct repository access using Sonnet 4 and 86.3 ± 1.1% using Sonnet 4.6. Query cost and latency were lower for Paper2Agent. Across 42 execution-based tasks from 10 non-biology computational papers, it achieved 98.1 ± 0.8% accuracy across five independent runs. Across 26 data- and discovery-focused papers, the resource layer achieved 89.0 ± 3.1% accuracy on 100 synthesis questions, outperforming a browser-use baseline while being cheaper and faster. The system correctly rejected 100% of out-of-scope queries in a permuted paper-question benchmark.