MCP Threat Modeling and Tool Poisoning

Capture Summary

Paper threat-modeling MCP implementations with STRIDE and DREAD across MCP Host/Client, LLM, MCP Server, external data stores, and authorization server. It focuses empirical evaluation on tool poisoning across major MCP clients.

Relevance

• Useful for converting MCP risks into structured threat models.
• Supports research on client-side validation, model decision path tracking, anomaly detection, and user transparency.

Collection Notes

Collected as current MCP threat-modeling source.