Tool Poisoning in MCP-Connected Agents: A 2026 Threat Model
Model Context Protocol turned every internal API into an LLM-reachable tool. The threat model didn’t catch up. Here’s what tool poisoning looks like in production and how to harden against it.