Security Robustness on MCPSafetybench
5ASR (w/o)MCPShield
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MCPShieldBackbone=Gemini3-Flash2026.02 | 5 | 100 | 100 | 100 | |
| MCPShieldBackbone=Kimi-K22026.02 | 5 | 84.44 | 83.33 | 88.89 | |
| MCPShieldBackbone=Qwen3 235B2026.02 | 5 | 100 | 100 | 100 | |
| MCPShieldBackbone=Deepseek V3.22026.02 | 10 | 100 | 100 | 100 | |
| MCPShieldBackbone=GPT4o-mini2026.02 | 15 | 94.44 | 94.44 | 94.44 | |
| MCPShieldBackbone=Minimax-M22026.02 | 35 | 84.04 | 66.67 | 100 |