Research
MCP Pitfall Lab: Exposing Developer Pitfalls in MCP Tool Server Security under Multi-Vector Attacks
Publication date not yet evaluated · Added arxiv.org
Page published
Coverage timeline
Single-source research — one report is available.
Why it matters
MCP Pitfall Lab quantifies how ordinary developer choices in tool exposure, schema design, and validation leave agentic LLM tool servers open to prompt injection, cross-tool forwarding, and image-to-tool leakage, giving defenders a measurable framework to audit and harden their MCP supply chain.
MCP Pitfall Lab is a protocol-aware security testing framework that models developer pitfalls in Model Context Protocol tool servers as reproducible scenarios and validates outcomes with MCP traces rather than agent self-report. Across 2,579 validator-completed runs over four models, it observed a 31.9% overall attack success rate (with multi-modal injection strongest at 38.7%) and introduces a Semantic MCP Bill-of-Materials to inventory tool semantics, trust boundaries, and audit support for hardening.