MCP Supply Chain Attacks: Tool Poisoning AI Agents

How malicious MCP servers use tool-description poisoning and split-instruction injection to hijack coding agents. Abstract The Model Context Protocol gave AI coding assistants a standard way to call