Investigate
Risk Intelligence Matrix
Tool risk prioritization, threat severity scoring, active guardrail signatures, and risk mitigation relationships.
MCP Tool Risk Classification & Severity Scores
Baseline risk assessment and required security controls per registered tool
Financial execution tool requiring strict capability validation and double-nonce verification.
Destructive database operation. Protected by saga rollback and zero-trust policy guardrails.
Content injection detection active via local regex and AWS Bedrock fallback guardrails.
Read-only telemetry and lookup operations with minimal security risk surface.
Active Content Guardrail Signatures
Deterministic inspection rules evaluated by LocalDemoGuardrailProvider prior to tool execution
| Detection Pattern | Threat Vector | Severity | Inspected Tools |
|---|---|---|---|
| /drop\s+table/i | Destructive SQL Injection | CRITICAL | create_user_record, send_email |
| /ignore\s+previous\s+instructions/i | Direct Prompt Injection Override | CRITICAL | create_user_record, send_email |
| /<script.*?>/i | Cross-Site Scripting / Payload Exfiltration | HIGH | create_user_record, send_email |
| /delete\s+all\s+records/i | Unauthorized Bulk Destruction Intent | HIGH | create_user_record, send_email |
| /system\s+override/i | Privilege Escalation Command | HIGH | create_user_record, send_email |
| /exploit/i | Adversarial Exploit Marker | MEDIUM | create_user_record, send_email |
Risk-to-Control Relationships
How SentinelMCP maps threat severity to cryptographic and transactional defenses
execute_transfer enforces constant-time HMAC-SHA256 verification, action match, amount ceiling checks, and atomic nonce persistence in SQLite.
delete_record captures the prior record snapshot before removal so reverse Saga rollbacks can restore deleted data if a multi-step plan fails.
No blocked tool attempts recorded yet in SQLite audit_events.