SAF-MCP에 대하여

Secure Agentic Framework(SAF)는 에이전트 시스템 내 적대자의 행동을 문서화합니다. SAF-MCP는 프로젝트 및 MCP 프로필로, 동일한 영구 기술 ID를 SAF Core, Code-Agent, RAG 및 Memory, Financial-Agent, Model-Lifecycle 프로필에도 적용할 수 있습니다. Framework Model v2는 인가, 라이프사이클, 관계, 릴리스 규칙을 정의합니다.

주요 특징

• 원자적이고 영구적인 기술: ID는 서로 다른 적대 행위를 나타내며, 항목이 통합되거나 폐기되더라도 영구적으로 유지됩니다.
• 프로파일 기반 범위: SAF는 일반적인 에이전트 메커니즘을 MCP 및 기타 도메인 특화 프로파일과 분리하여 기술 ID를 중복하지 않습니다.

SAF-MCP: Secure Agentic Framework for Model Context Protocol

About SAF-MCP

The Secure Agentic Framework (SAF) documents adversary behavior in agentic systems. SAF-MCP is the project and MCP profile: the same permanent technique IDs can also be scoped to SAF Core, Code-Agent, RAG and Memory, Financial-Agent, and Model-Lifecycle profiles. Framework Model v2 defines the admission, lifecycle, relationship, and release rules.

Key Features

• Atomic, Permanent Techniques: IDs represent distinct adversary behaviors and remain permanent even when an entry is consolidated or deprecated.
• Profile-Based Scope: SAF separates general agentic mechanisms from MCP and other domain-specific profiles without duplicating technique IDs.
• Typed Relationships: Technique links distinguish specializations, prerequisites, sequence, overlap, alternatives, and replacements.
• MITRE ATT&CK Alignment: Where defensible, SAF techniques link to relevant MITRE ATT&CK behaviors to support threat modeling, detection engineering, and cross-framework analysis. These mappings do not establish compliance.
• Evidence-Led Releases: Evidence, taxonomy, and operational gates require traceable claims, coherent placement, and tested or explicitly bounded detection.
• Operational Coverage Registry: The generated detection coverage matrix separates technique linkage, observation modality, validation maturity, and external rule ownership.
• Actionable Mitigations: Each technique includes detailed mitigation strategies and detection rules to help defenders protect their MCP deployments.

How to Use This Framework

1. Security Teams: Select the profiles that match your agentic deployment, then use the catalog below to build a threat model
2. Developers: Review techniques relevant to the agents, models, data paths, and tools you operate and implement the linked mitigations
3. Compliance Officers: Map SAF-MCP techniques to your existing security controls via MITRE ATT&CK linkages
4. Red Teams: Reference attack techniques for security testing of MCP deployments

TTP Reference Table