675 MCP Servers Scored on Security: 89% Fail, Only 6 Deemed Safe
Low_Location1261 · reddit · 2026-09-04
RepoAI scored all 675 published MCP servers in its directory on 15 structural security signals (read-only mode, auth, maintenance, publisher). Results: only 0.9% rated safe, 9.8% medium, 89.3% high risk.
Key findings
- 92.1% offer no read-only mode — installs get full read/write access
- 35.9% ship dangerous tools (code execution, deletion, file writes); 43.1% of those have no read-only escape hatch
- Only 14.7% support OAuth; 1.9% have no authentication at all
- Only 12% are published by the actual vendor
Ruling out selection bias: 20 popular unindexed GitHub MCP servers (top: 1,860 stars) were scored cold — none reached medium tier, best was 55/100.
Case study: mcp-server-trello exposes 35 tools including dangerous delete operations with no read-only mode, scoring 20/100.
The authors invoke Log4Shell (dormant for 8 years) to argue absence of incidents isn't evidence of safety, and highlight MCP's specific prompt-injection attack surface: an assistant needs no server bug to be tricked into abusing its permissions.
More from coding & agent
- PostgreSQL 19 slips to November as LLM-assisted code review floods committers with edge cases — sanjaykalra · 2026-10-21
- Open-sourced pixel art normals generator ships as Rust CLI and wasm library — HugoDuprez · 2026-10-20
- Eval researcher teases guide to subjective-task evals, with a curated resource list — cwolferesearch · 2026-10-20
- Eon cuts engineer onboarding from 2 days to 2 minutes with an MCP Gateway on Bedrock AgentCore — AWS ML Blog · 2026-10-20
- On building persistent agents: owning your agent's memory stack — omarsar0 · 2026-10-20
- Mintlify launches Autopilot beta: an always-on agent that keeps docs in sync with your code — ycombinator · 2026-10-20