MCP directory self-audit: 704 of 711 'serious' security findings were false positives
Glittering_Device653 · reddit · 2026-11-17
The operator of HostedMCP (tracking 123k MCP servers and 71k agent skills with automated checks) audited their own published findings and found 704 of 711 serious security findings were wrong.
Root causes: the "hidden instructions" rule matched words, flagging security tools that quote "ignore previous instructions" as a warning; "Never disable security headers" counted as disabling security; typosquat checks called mssql a mysql look-alike; destructive-tool checks failed tools for a missing destructiveHint (true by default per spec); 46% of "no license" findings actually had license files GitHub couldn't match to an SPDX id.
The fix: rules now report instructions, not words — quoted text, code blocks and "never do this" lists are excluded. Replaying the same 711 findings leaves 7, hand-verified: six scripts piping installers into sh, and one skill telling the agent to switch a check off. The author admits the stricter rules now miss things, like data-exfiltrating scripts that look like ordinary API calls. Maintainers can claim listings at hostedmcp.org and correct errors.
More from coding & agent
- Sigvard AI launches automated LLM fine-tuning platform with GPU selection, seeks feedback — BeatedBull · 2026-11-17
- AI horror short 'The Day She Arrived' built with Midjourney, Hailuo H3 and ElevenLabs — Kyrannio · 2026-11-17
- Developer ships ASC CLI 7.0.0 ahead of schedule — rudrank · 2026-11-17
- Grok Bot Content System: Automate Research, Never Judgment — a $10K Side-Biz Play — PrajwalTomar_ · 2026-11-17
- Qwen Code Desktop v0.28.0 ships DingTalk managed-agent adapter and workspace sidebar filter — github-actions[bot] · 2026-11-17
- Asking an AI to use verifiable info changes its picks dramatically — santiviquez · 2026-11-17