🔬 Security Standards

MCP Security Assessment Standard v1.0

CraftedTrust's comprehensive security assessment framework, aligned with the Coalition for Secure AI (CoSAI) MCP Security Framework and the OWASP Top 10 for Agentic Applications.

12-Category Assessment Framework

Every MCP server is assessed across 12 weighted categories totaling 100 points. Categories are organized into five security domains.

Authentication & Access
01 10 pts
Identity & Auth
OAuth detection, auth method analysis, credential exposure assessment
Authentication & Access
02 8 pts
Permission Scope
Tool permission analysis, least-privilege assessment, dangerous parameter detection
Authentication & Access
Server Security
03 8 pts
Transport Security
TLS/HTTPS enforcement, certificate validation, HSTS headers
Server Security
04 10 pts
Network Behavior
Outbound connection analysis, undeclared endpoint detection, suspicious domain identification
Server Security
05 8 pts
Protocol Compliance
MCP spec version validation, JSON-RPC compliance, required field verification
Server Security
Tool Safety
06 8 pts
Declaration Accuracy
Declared vs actual capability matching, description completeness
Tool Safety
07 10 pts
Tool Integrity
Hidden instruction detection, tool name shadowing, rug-pull change detection
Tool Safety
08 8 pts
Input Validation
Schema validation presence, type checking, injection resistance indicators
Tool Safety
Supply Chain
09 8 pts
Supply Chain
Package provenance, dependency freshness, known CVE detection
Supply Chain
10 6 pts
Code Transparency
Open source availability, repository activity, license clarity
Supply Chain
11 8 pts
Publisher Trust
Publisher verification status, history, reputation signals
Supply Chain
Data Handling
12 8 pts
Data Protection
PII exposure detection, data flow analysis, cache control headers
Data Handling

CoSAI Framework Mapping

How CraftedTrust's 12 assessment categories map to the Coalition for Secure AI (CoSAI) threat categories.

CraftedTrust Category CoSAI Threat Category Coverage
Identity & Auth Identity & Access Management Full
Permission Scope Overprivileged Tokens Full
Transport Security Transport Security Full
Network Behavior System Integrity Full
Protocol Compliance Protocol-level Security Full
Declaration Accuracy Confused Deputy Partial
Tool Integrity Tool Poisoning Full
Input Validation Input Validation Full
Supply Chain Supply Chain Security Full
Code Transparency Supply Chain Security Partial
Publisher Trust Identity & Access Management Partial
Data Protection Data Protection Full

Mapping based on CoSAI Agentic AI Security Framework. CraftedTrust categories are designed to comprehensively cover the threat landscape identified by the Coalition for Secure AI.

OWASP Top 10 for Agentic Applications Mapping

How CraftedTrust's assessment categories address risks identified in the OWASP Top 10 for Agentic Applications.

OWASP Agentic Risk CraftedTrust Categories Coverage
A01: Prompt Injection Tool Integrity, Input Validation Full
A02: Improper Tool/Plugin Use Declaration Accuracy, Permission Scope Full
A03: Excessive Agency Permission Scope, Network Behavior Full
A04: Insecure Output Handling Data Protection, Input Validation Partial
A05: Overreliance on AI Output (Informational — outside scan scope) N/A
A06: Sensitive Information Disclosure Data Protection, Identity & Auth Full
A07: Insecure Plugin Design Tool Integrity, Protocol Compliance Full
A08: Supply Chain Vulnerabilities Supply Chain, Code Transparency Full
A09: Excessive Permissions Permission Scope, Identity & Auth Full
A10: Insufficient Logging/Monitoring (Covered by CraftedTrust Audit Logger) External

EU AI Act Compliance Mapping (Articles 9–15)

How CraftedTrust's assessment maps to EU AI Act requirements for high-risk AI systems.

EU AI Act Article Title CraftedTrust Categories Coverage
Art. 9 Risk Management Network Behavior, Tool Integrity, Input Validation, Data Protection Full
Art. 10 Data Governance Data Protection, Supply Chain, Code Transparency Full
Art. 11 Technical Documentation Declaration Accuracy, Code Transparency, Protocol Compliance Full
Art. 12 Record-Keeping Protocol Compliance, Network Behavior (+ Audit Logger) Full
Art. 13 Transparency Declaration Accuracy, Code Transparency, Publisher Trust Full
Art. 14 Human Oversight Identity & Auth, Permission Scope (+ Audit Logger approve mode) Full
Art. 15 Accuracy, Robustness & Cybersecurity Transport Security, Input Validation, Tool Integrity, Identity & Auth, Network Behavior Full

CraftedTrust's Audit Logger provides built-in compliance with Article 12 (record-keeping) and Article 14 (human oversight) through its approve mode feature.

NIST AI RMF Mapping

How CraftedTrust's assessment categories map to the NIST AI Risk Management Framework functions and subcategories.

NIST AI RMF Function Subcategories Covered CraftedTrust Categories Coverage
GOVERN GV-1.3 Transparency, GV-1.6 Accountability, GV-3.2 Policies Code Transparency, Publisher Trust, Declaration Accuracy Full
MAP MP-2.3 Risk Identification, MP-3.4 Dependencies, MP-4.2 Impact Assessment Supply Chain, Network Behavior, Permission Scope, Data Protection Full
MEASURE MS-1.1 Security Testing, MS-2.3 Validation, MS-2.6 Monitoring Tool Integrity, Input Validation, Protocol Compliance, Transport Security Full
MANAGE MG-2.2 Incident Response, MG-2.4 Risk Treatment, MG-3.1 Access Control Identity & Auth, Permission Scope, Network Behavior, Data Protection Full

AIUC-1 Risk Domain Mapping

How CraftedTrust's assessment categories map to the AIUC-1 AI agent security standard risk domains.

AIUC-1 Domain Key Controls CraftedTrust Categories Coverage
A: Agent Identity & Authentication Agent credential management, mutual authentication, identity lifecycle Identity & Auth, Publisher Trust Full
B: Authorization & Access Control Least-privilege enforcement, dynamic permission boundaries, scope limitations Permission Scope, Identity & Auth Full
C: Data Protection & Privacy Data flow controls, PII handling, encryption requirements, data minimization Data Protection, Transport Security Full
D: Tool & Plugin Security Tool validation, input sanitization, capability declaration, integrity checks Tool Integrity, Input Validation, Declaration Accuracy Full
E: Supply Chain & Dependency Management Dependency verification, provenance tracking, vulnerability scanning Supply Chain, Code Transparency Full
F: Monitoring & Incident Response Runtime monitoring, anomaly detection, audit logging, incident workflows Network Behavior, Protocol Compliance Full

AIUC-1 is the world's first AI agent security standard, developed with 100+ Fortune 500 CISOs. CraftedTrust's 12-factor assessment provides direct coverage across all six AIUC-1 risk domains.

Scoring & Grading Methodology

The total trust score ranges from 0 to 100, computed from 12 weighted categories. Scores map to letter grades and trust labels.

Letter Grades

A
90 – 100
B
75 – 89
C
60 – 74
D
40 – 59
F
0 – 39

Trust Labels

Trusted
80+
Moderate
60 – 79
Caution
40 – 59
Warning
20 – 39
Dangerous
0 – 19

Scan Depth Tiers

Each scan depth tier builds on the previous, adding progressively deeper security analysis.

Standard
Free
Automated 12-factor scan with basic checks across all categories. Includes transport security verification, tool discovery, and fundamental compliance validation.
Enhanced
Standard Certification
Everything in Standard, plus protocol compliance probes, error handling tests, and deeper permission analysis with edge-case validation.
Deep
Premium / Enterprise
Everything in Enhanced, plus full tool validation, response consistency checks, header security analysis, and behavioral anomaly detection.
Certification
Deep Scan + Compliance Mapping
Full Deep scan plus CoSAI framework mapping, OWASP agentic risk mapping, EU AI Act compliance mapping, NIST AI RMF mapping, AIUC-1 risk domain mapping, and a comprehensive PDF assessment report for compliance and audit purposes.

Resources

Further reading on the frameworks and tools referenced in this standard.