Skip to content

Latest commit

 

History

History
158 lines (128 loc) · 9.27 KB

File metadata and controls

158 lines (128 loc) · 9.27 KB

SAF-MCP Mitigations Reference

About SAF-MCP Mitigations

SAF-MCP mitigations are security controls designed to protect Model Context Protocol (MCP) implementations from the attack techniques documented in our framework. Each mitigation is categorized by type and effectiveness, with clear mappings to the techniques it addresses.

Licensing

New contributions to the mitigations are licensed under the Community Specification License 1.0. Mitigation content contributed on or before 2026-06-10 remains under CC BY 4.0 until the original contributors sign off on relicensing or the content is rewritten. See LICENSE for the full licensing structure.

Mitigation Categories

  • Architectural Defense: Fundamental design patterns that prevent entire classes of attacks
  • Cryptographic Control: Security measures using cryptographic techniques
  • AI-Based Defense: Controls leveraging AI/ML for detection and prevention
  • Input Validation: Sanitization and validation of inputs before processing
  • Supply Chain Security: Controls for securing the MCP software supply chain
  • UI Security: Controls ensuring visual consistency and preventing deception
  • Isolation and Containment: Sandboxing and isolation techniques
  • Detective Control: Monitoring and detection capabilities
  • Preventive Control: Controls that prevent attacks before they occur
  • Architectural Control: System design patterns for security

Effectiveness Ratings

  • High: Highly effective control, prevents 80%+ of targeted attacks
  • Medium-High: Effective control, prevents 60-80% of targeted attacks
  • Medium: Moderately effective, prevents 40-60% of targeted attacks
  • Low: Limited effectiveness, prevents <40% of targeted attacks

Mitigation Overview

Mitigation ID Name Category Effectiveness
SAF-M-1 Control/Data Flow Separation Architectural Defense High (Provable Security)
SAF-M-2 Cryptographic Integrity for Tool Descriptions Cryptographic Control High
SAF-M-3 AI-Powered Content Analysis AI-Based Defense Medium-High
SAF-M-4 Unicode Sanitization and Filtering Input Validation Medium-High
SAF-M-5 Content Sanitization Input Validation Medium
SAF-M-6 Tool Registry Verification Supply Chain Security High
SAF-M-7 Content Rendering Parity UI Security Medium-High
SAF-M-8 Visual Validation UI Security Medium
SAF-M-9 Sandboxed Testing Isolation and Containment High
SAF-M-74 Per-Invocation Capability Brokering Isolation and Containment High
SAF-M-10 Automated Scanning Detective Control Medium
SAF-M-11 Behavioral Monitoring Detective Control High
SAF-M-12 Audit Logging Detective Control Medium-High
SAF-M-13 OAuth Flow Verification Preventive Control High
SAF-M-14 Server Allowlisting Preventive Control High
SAF-M-15 User Warning Systems Preventive Control Medium
SAF-M-16 Token Scope Limiting Preventive Control High
SAF-M-17 Callback URL Restrictions Preventive Control High
SAF-M-18 OAuth Flow Monitoring Detective Control Medium
SAF-M-19 Token Usage Tracking Detective Control Medium
SAF-M-20 Anomaly Detection Detective Control High
SAF-M-21 Output Context Isolation Architectural Control High
SAF-M-22 Semantic Output Validation Input Validation Medium-High
SAF-M-23 Tool Output Truncation Preventive Control Medium
SAF-M-24 SBOM Generation and Verification Supply Chain Security High
SAF-M-25 AI-Specific Risk Modeling Risk Management Medium-High
SAF-M-26 Data Provenance Tracking Data Security High
SAF-M-27 Social Engineering Awareness Training Human Factors Medium
SAF-M-28 Pre-Authentication Tool Concealment Preventive Control High
SAF-M-30 Vector Store Integrity Verification Cryptographic Control High
SAF-M-32 Continuous Vector Store Monitoring Detective Control Medium-High
SAF-M-33 Training Data Provenance Verification Data Security High
SAF-M-34 AI Model Integrity Validation Cryptographic Control High
SAF-M-35 Adversarial Training Data Detection AI-Based Defense Medium-High
SAF-M-36 Model Behavior Monitoring Detective Control Medium-High
SAF-M-29 Explicit Privilege Boundaries Architectural Control High
SAF-M-37 Metadata Sanitization Input Validation High
SAF-M-38 Schema Validation Input Validation Medium-High
SAF-M-39 Prompt Context Validation Architectural Control High
SAF-M-40 Clear UI Patterns UI Security Medium-High
SAF-M-41 Tool and Package Pinning Supply Chain Control High
SAF-M-42 Cross-Server Protection Architectural Control High
SAF-M-43 Steganography Scanner Detective Control Medium-High
SAF-M-44 Behavioural Monitoring Detective Control Medium-High
SAF-M-45 Tool Manifest Signing & Server Attestation Supply Chain Security High
SAF-M-46 Bridge Risk Management Preventive Control High
SAF-M-47 Cross-Chain Transaction Graph Analysis Detective Control High
SAF-M-48 Custodial Off-Ramp Monitoring Detective Control Medium-High

Summary Statistics

  • Total Mitigations: 48
  • High Effectiveness: 27 (56%)
  • Medium-High Effectiveness: 15 (31%)
  • Medium Effectiveness: 6 (13%)
  • Low Effectiveness: 0 (0%)

Category Distribution

Category Number of Mitigations
Detective Control 12
Preventive Control 7
Input Validation 6
Cryptographic Control 3
Architectural Defense 2
UI Security 3
AI-Based Defense 2
Supply Chain Security 4
Data Security 2
Architectural Control 4
Isolation and Containment 2
Risk Management 1
Human Factors 1

Implementation Guidance

Defense in Depth Strategy

The most effective security posture combines multiple mitigations across different categories:

  1. Foundation Layer: Implement architectural defenses (SAF-M-1, SAF-M-21) that provide fundamental protection
  2. Prevention Layer: Add cryptographic controls (SAF-M-2) and input validation (SAF-M-4, SAF-M-5, SAF-M-22)
  3. Detection Layer: Deploy monitoring and detection controls (SAF-M-10, SAF-M-11, SAF-M-12)
  4. Response Layer: Maintain audit logs and incident response procedures

Priority Implementation

For organizations with limited resources, prioritize implementation based on:

  1. Critical Controls (Implement First):

    • SAF-M-1: Control/Data Flow Separation
    • SAF-M-2: Cryptographic Integrity
    • SAF-M-6: Tool Registry Verification
    • SAF-M-11: Behavioral Monitoring
  2. Important Controls (Implement Second):

    • SAF-M-3: AI-Powered Content Analysis
    • SAF-M-4: Unicode Sanitization
    • SAF-M-9: Sandboxed Testing
    • SAF-M-13: OAuth Flow Verification
  3. Additional Controls (Implement as Resources Allow):

    • Remaining mitigations based on specific threat model

Usage Guidelines

  • Review mitigations relevant to your threat model
  • Implement controls in layers for defense in depth
  • Regularly update and test mitigation effectiveness
  • Monitor for new threats requiring additional controls
  • Consider automation for detective controls
  • Document implementation details for compliance

Contributing

To add new mitigations or update existing ones:

  1. Create a new directory under mitigations/ with the next available SAF-M-X number
  2. Use the mitigation template for consistent documentation
  3. Update this MITIGATIONS.md file
  4. Submit a pull request with justification for the new mitigation