Threat Coverage
How our detection rules map to each risk in the MCP Top 10. A risk is covered once at least one rule is live; partial means rules are in review.
The six threat categories from the official Model Context Protocol security best practices, mapped 1:1 to the MCPSafe rules that cover them. ✅ covered, ⚠️ in review, ❌ missing.
Tool Abuse
Tools used beyond their declared intent or scope
Prompt Injection
Untrusted input redirecting agent behaviour
Insufficient Access Control
Missing auth, wildcard permissions, or over-privileged roles
LLM Prompt Manipulation
Injecting instructions into inner-LLM system prompts
Data Exfiltration
Sensitive data leaking through tool responses or logs
Insecure Defaults
Dangerous configuration shipped as the default
Supply-Chain Compromise
Malicious code introduced via dependencies or install hooks
Secrets Exposure
Credentials or keys stored or transmitted in plaintext
Rule statuses update on every build from the live rules catalog. Rules in review do not yet affect user-visible scan results.