Security Robustness on MCPSecbench
0ASR (w/o Config)MCPShield
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MCPShieldBackbone=GPT4o-mini2026.02 | 0 | 96.67 | 91.67 | 100 | |
| MCPShieldBackbone=Kimi-K22026.02 | 0 | 100 | 100 | 100 | |
| MCPShieldBackbone=Deepseek V3.22026.02 | 0 | 100 | 100 | 100 | |
| MCPShieldBackbone=Qwen3 235B2026.02 | 0 | 96.67 | 91.67 | 100 | |
| MCPShieldBackbone=Gemini3-Flash2026.02 | 8.33 | 93.33 | 91.67 | 100 | |
| MCPShieldBackbone=Minimax-M22026.02 | 8.33 | 85.45 | 81.82 | 90.91 |