Malicious server detection on Benign MCP Servers
29.41Average ScoreMCPShield
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MCPShieldBackbone=Qwen3-235B2026.02 | 29.41 | 17.65 | 35.29 | 0.0052 | |
| MCPShieldBackbone=Deepseek V3.22026.02 | 18.82 | 11.76 | 23.53 | 0.0024 | |
| MCPShieldBackbone=Minimax-M22026.02 | 12.06 | 0 | 23.53 | 0.0089 | |
| MCPShieldBackbone=Kimi-K22026.02 | 3.53 | 0 | 11.76 | 0.0028 | |
| MCPShieldBackbone=Gemini3-Flash2026.02 | 3.21 | 0 | 8.33 | 0.0019 | |
| MCPShieldBackbone=GPT 4o-mini2026.02 | 2.35 | 0 | 5.88 | 0.001 |