MCPShield: Content-Aware Attack Detection for LLM Agent Tool-Call Traffic

TL;DR

MCPShield introduces a graph-based attack detection framework for LLM agent tool-call traffic, emphasizing content features and revealing significant evaluation pitfalls.

Contribution

It presents a novel graph neural network approach for detecting attacks in MCP tool-call traffic, highlighting the importance of content embeddings and evaluation methodology.

Findings

Content embeddings significantly improve detection AUROC above 0.89.

Naive random-split evaluation inflates AUROC by up to 26 percentage points.

Tree ensembles on pooled embeddings achieve AUROC of 0.975, outperforming neural models.

Abstract

The Model Context Protocol (MCP) has become a widely adopted interface for LLM agents to invoke external tools, yet learned monitoring of MCP tool-call traffic remains underexplored. In this article, MCPShield is presented as an attack detection framework for MCP tool-call traffic that encodes each agent session as a graph (tool calls as nodes, sequential and data-flow links as edges), enriches nodes with sentence-embedding features over arguments and responses, and classifies sessions as benign or attacked. Three GNN architectures (GAT, GCN, GraphSAGE), a no-graph MLP, and classical baselines (XGBoost, random forest, logistic regression, linear SVM) are evaluated, with the full architecture comparison conducted on RAS-Eval (task-stratified splits) and GraphSAGE retained as the GNN baseline on ATBench and a combined-source variant (both label-stratified). Three findings emerge. First, content-level features are essential: metadata-only detection plateaus around an AUROC of 0.64 regardless of architecture, while content embeddings push the AUROC above 0.89. Second, naive random-split evaluation inflates AUROC by up to 26 percentage points relative to task-disjoint splits, a memorization confound that prior agent-detection work has not addressed. Third, the detection signal resides primarily in the SBERT content embeddings: an AUROC of 0.975 was reached by tree ensembles on pooled embeddings, performing, for the most part, better than the neural architectures in the primary RAS-Eval setting including GNNs (0.917) and the MLP (0.896), and self-supervised pre-training does not deliver a label-efficiency advantage on this task.