The MCPTrust scoring model
Base 50. Verifiable good posture adds points. Verified bad posture subtracts. Entries that declare almost nothing stay low — that is intentional: absence of evidence is not evidence of safety.
| Signal | Points | Why |
|---|---|---|
| Public repository | +15 | Source is auditable. |
| License file | +10 | Legal status of the code is explicit. |
| SECURITY.md | +5 | Disclosed vulnerability-report channel. |
| OAuth metadata published | +10 | Proper auth discovery instead of raw bearer tokens. |
| Repository >90 days old | +10 | Track record. |
| Pushed in last 90 days | +10 | Actively maintained in a fast-moving ecosystem. |
| Organization-owned | +5 | Continuity signal. |
| All remotes reachable | +5 | The service is actually up. |
| Plain-HTTP remote | −40 | Unencrypted tool traffic. |
| Registry status ≠ active | −50 | Officially deprecated/deleted. |
| All remotes unreachable | −15 | Dead or sleeping service. |
| Repo stale >180 days | −15 | Abandonment risk. |
Grades: ≥80 Verified · 60–79 Standard · 40–59 Caution · <40 High Risk.
Scope & limits: we grade the auditable posture, not the code itself. A perfect score does not mean "no vulnerabilities" — it means "everything that can be checked publicly, checks out". Closed-source commercial servers start at 50 by design. Scan batch #1: first 60 latest-version entries from the official MCP registry.
Scope & limits: we grade the auditable posture, not the code itself. A perfect score does not mean "no vulnerabilities" — it means "everything that can be checked publicly, checks out". Closed-source commercial servers start at 50 by design. Scan batch #1: first 60 latest-version entries from the official MCP registry.