MCP’s biggest security loophole is identity fragmentation (www.techradar.com)

Model Context Protocol (MCP is enabling LLMs to talk to each other, tools and private data sources, and its adoption is already large (Backslash Security cites ~15,000 MCP servers). But recent audits and vendor reports (including Trend Micro) show hundreds of MCP endpoints leaking sensitive data and enabling remote code execution, often because of hardcoded or over-privileged static credentials and incomplete access controls. The core point: attackers rarely exploit a protocol itself—they exploit mistakes in identity and credential management, and MCP’s lack of built-in identity guarantees makes it an attractive vector when identities are spoofed. The article argues the fundamental remedy is identity-first security, not protocol patching. MCP deployments need strong, cryptographic authentication (TPMs, hardware/bio-backed keys), ephemeral credentials and just‑in‑time privileges, and a single source of truth for identity and access that spans humans, machines and AI agents. Tying LLM access controls to the enterprise identity system and avoiding siloed “AI-only” identity solutions reduces role bloat, standing privileges and impersonation risk. In short, securing MCP is about eliminating fragmented, static identities so AI and protocols operate under consistent, verifiable identity and least-privilege policies.