Securing the Model Context Protocol: Defending LLMs against Tool Poisoning and Adversarial Attacks

The Model Context Protocol MCP enables Large Language Models to integrate external tools through structured descriptors, increasing autonomy in decision-making, task execution, and multi-agent workflows. However, this autonomy creates a largely overlooked security gap. Existing defenses focus on...