Anatomy of MCP Security: 8 Critical Risks and 40% Unauthenticated Servers
mnelyzeaN · reddit · 2026-07-30
Security team Hacken analyzed the 8 core security risks of the Model Context Protocol (MCP), noting that OAuth and least privilege are far from sufficient.
Key Risks Include:
- Prompt Injection Escalation: In an MCP context, injection tricks the model into executing dangerous tool calls rather than just generating bad text.
- Overpowered Tools: The blast radius is determined by the tool's capabilities (e.g., shell access) if the model is tricked into calling it.
- Zero Authentication: A scan revealed that 40.55% of live remote MCP servers expose tools without any authentication, leaking internal data.
- OAuth Flaws: 100% of 119 OAuth-enabled servers tested had vulnerabilities (325 total flaws), with 96.6% having dynamic client registration issues.
- Token Passthrough & Context Poisoning: Cross-service token abuse and tampered document resources can stealthily steer the model's execution logic.
- Local Server Compromise: Local MCP servers with unrestricted access to home directories bridge supply chain risks directly to agent vulnerabilities.
Related event: Research Reveals 8 Major Security Risks in MCP Protocol(2 posts)→
More from coding & agent
- Reflection's Beam, a 501B-parameter open MoE model with 1M context, lands on Hugging Face under Apache 2.0 — QGallouedec · 2026-10-16
- 'LLM code is sloppy' rebutted with real ffmpeg code: identical if/else branches in a top library — kingslyj · 2026-10-16
- Office-editing tool calls need file, worktree and unit: lessons from dsh-univer-office — FunDistribution6537 · 2026-10-16
- Architect MD workflow: Gemini one-shot outputs that beat Opus, via 5-lens plan critique — oikk01 · 2026-10-16
- Prove a coding agent actually ran the tests: check artifacts, not approvals — ComprehensiveShake76 · 2026-10-16
- Building an acceptance harness for AI spreadsheet edits: allowlists, recalc checks, idempotency — ukdicha-modakkk · 2026-10-16