Paper2Agent – transforming research papers into interactive AI agents (github.com)

Paper2Agent is a multi-agent pipeline that automatically turns a research paper’s codebase into an interactive AI agent with minimal human input. Given a GitHub URL and a project directory, the tool scans and runs tutorials/notebooks, extracts reusable “tools” from tutorial code, spins up an isolated Python environment, and generates a Claude-compatible MCP server (src/<repo>_mcp.py) and tool modules (src/tools/). It integrates with Claude Code (or Google Gemini CLI), supports tutorial filtering, authenticated repos via API keys, and provides shell commands to build and launch agents. Typical processing takes 30 minutes to 3+ hours and—using Claude Sonnet 4—costs roughly $15 for complex repos like AlphaGenome. Output artifacts include executed notebooks, test suites, environment manifests, and JSON reports documenting discovery, execution, and tool extraction. For AI/ML researchers the significance is twofold: it automates reproducibility and turns static papers into conversational, actionable assistants that can run analyses (examples: TISSUE, Scanpy, AlphaGenome) and answer domain queries like calling causal genes or computing prediction intervals. Technically, Paper2Agent standardizes tool extraction, environment isolation, and MCP server creation, lowering friction for adopting published methods. Practical caveats include LLM dependence for orchestration, the need to configure runtime/authentication, and potential correctness risks from automated execution—nonetheless it promises a faster pathway from paper to interactive, testable research tooling.