Stanford's Paper2Agent turns 74 of 100 bio papers to agents

Stanford researchers have built Paper2Agent, a framework that wraps a scientific paper's text, code and data into an AI agent that can answer questions about the work and rerun its methods on new inputs. In the team's own tests, described in *Nature* on 16 September 2026 by Kaia Glickman, the system converted 74 of 100 computational biology papers into working agents.

The framework, built by postdoctoral scholar Jiacheng Miao, Associate Professor James Zou and colleagues at Stanford, exposes each paper as a Model Context Protocol server that a chat agent like Claude Code can call. Across the 74 converted papers, the 593 validated tools averaged 91.2% accuracy on 300 benchmark questions at roughly $0.20 per query. An agent built from the AlphaGenome paper carried 22 MCP tools and was assembled in about 45 minutes for around $14.

Zou frames the shift as more than a reproducibility aid. "This is an opportunity to fundamentally reimagine what knowledge looks like," he tells Nature. "Instead of having only passive artifacts, why don't we convert each static record into an active embodiment of knowledge?" The team describes each converted output as a "virtual corresponding author": a live front door to the manuscript, supplementary materials, datasets and code, rather than a static PDF.

Zou also argues the conversion doubles as a quality signal. "Agentification itself is a useful certificate that says, 'This work is relatively complete and well documented.'" The counterpart to that certificate is the 26 papers that failed to convert in the biology sample, which points at published computational work that ships without the runnable code needed to be turned into a service.