Paper2Agent turns papers into executable MCP agents: 22 validated tools in 45 minutes, 100% on novel benchmarks

bravo_abad · x · 2026-09-17

Paper2Agent automates the jump from reading a method to actually using it — a paper tells you what was done, but reuse still means finding code, installing deps, and adapting the workflow.

It takes the paper, supplementary material, data, and codebase, then builds an MCP server with three components: executable tools derived from the original method, structured access to the paper's resources, and prompts encoding the scientific workflow. Crucially, generated tools are validated against results from the original code before the agent can use them.

For AlphaGenome, Paper2Agent produced 22 validated tools in 45 minutes; the resulting agent answered 100% of novel benchmark questions correctly vs. 78.7% for a Claude baseline.

Related event: Stanford's Paper2Agent, Published in Nature, Turns Papers into Interactive AI Agents(18 posts)→

Original post →

More from coding & agent

coding & agent channel →