MCP Security: Why Your AI Agent's Tool Calls Need...

TL;DR: The Model Context Protocol is becoming the standard way agents talk to tools. Most deployments have zero security between the agent and the tools it invokes. This post breaks down the real threat surface — tool poisoning, rug-pull attacks, data exfiltration through arguments — maps them to the OWASP Agentic Top 10, and lays out practical, implementable defenses with specific tools and thresholds. MCP in 60 Seconds The Model Context Protocol (MCP) is an open standard for connecting agents