Understanding Your Security Score

MCP Security Score provides multiple metrics to help you understand your MCP server's security posture. This guide explains each score and how to interpret them.

The Security Score

Your primary security score is a number from 0 to 100, representing the overall security health of your MCP server.

How It's Calculated

The score starts at 100 points and deducts points for each security finding:

| Severity | Points Deducted | |----------|-----------------| | Critical | -20 | | High | -12 | | Medium | -6 | | Low | -3 | | Info | -1 |

Diminishing Returns

To prevent a single category from tanking your entire score, we apply diminishing returns:

First 50 points deducted in a category: Full deduction
After 50 points: Further deductions are reduced by 50%

This ensures that even a repository with many low-severity findings won't automatically get an F.

Score Floor

The minimum score is 0. You cannot go negative.

Letter Grades

For quick assessment, scores map to letter grades:

| Score Range | Grade | Meaning | |-------------|-------|---------| | 90-100 | A | Excellent security posture | | 80-89 | B | Good, minor improvements possible | | 70-79 | C | Fair, some issues to address | | 60-69 | D | Poor, significant issues found | | 0-59 | F | Critical vulnerabilities present |

What Each Grade Means

Grade A (90-100)

No critical or high severity findings
Few medium or low findings
Well-structured, secure code
Ready for production use

Grade B (80-89)

No critical findings
May have a few high severity issues
Some medium/low findings to address
Generally safe, but improvements recommended

Grade C (70-79)

May have high severity findings
Several medium severity issues
Requires attention before production
Consider a security review

Grade D (60-69)

High severity vulnerabilities present
Many security issues across categories
Not recommended for production
Needs significant remediation

Grade F (0-59)

Critical vulnerabilities found
Serious security risks present
Do not deploy to production
Immediate action required

Safety Score

When AI analysis is enabled, you'll also see a Safety Score. This composite score combines:

60% - Static analysis score (the main security score)
40% - AI trust score (Claude's assessment of behavioral safety)

Why Two Scores?

Static analysis excels at finding known patterns but can miss:

Novel attack vectors
Semantic issues in code logic
Context-dependent vulnerabilities

AI analysis provides:

Behavioral understanding
Intent analysis
Risk assessment based on capabilities

Together, they give a more complete picture.

When They Diverge

If your static score and AI trust score differ significantly:

High static, low AI trust:

Code follows best practices but has risky capabilities
May have legitimate but dangerous operations
Review AI's specific concerns

Low static, high AI trust:

Many pattern matches but code is contextually safe
Could be false positives in test files
Review static findings for accuracy

Category Breakdown

Your score is broken down across security categories:

Categories and Weights

| Category | Weight | What It Covers | |----------|--------|----------------| | Tool Permissions | 25% | RCE risks, authentication | | Input Handling | 25% | Filesystem, data validation | | Dependencies | 20% | Supply chain, CVEs | | Code Patterns | 15% | Network, secrets | | Transparency | 15% | MCP-specific best practices |