Reproduce “MCP Tool Poisoning in Agentic AI Frameworks: A Cross-Framework Measurement Study” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Farooq Shaikh, Biagio Montaruli

The authors evaluate whether MCP frameworks protect agents from malicious tool metadata across 12 frameworks, 18 models, and 45 attack scenarios. Across 9,720 test combinations, 22.9% were vulnerable by dispatched canary-bearing calls or canary requests, not necessarily secret leakage. Nine frameworks had similar pooled rates, but this does not prove equivalence; results apply to tested configurations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.