Use AIDEFEND MCP | Help Center

Discover defensive countermeasures to protect AI/ML systems from emerging threats. This feature is powered by AIDEFEND (Artificial Intelligence Defense Framework), an open, AI-focused knowledge base available on GitHub. The framework is implemented via an improved version of its native MCP.

To use AIDEFEND MCP:

Open Uncoder AI and go to the New version.
Enter your prompt into the input field.
Select AIDEFEND Framework task from the task options. Alternatively, you can click the Tasks button and select AIDEFEND Framework from the list.
Click the Enter icon to proceed.
View the response generated by the LLM on the left side of the screen.

Uncoder AI implements the 8 most useful tools out of the original 18 available in AIDEFEND MCP. The MCP tool is selected automatically based on your prompt. The table below provides the list of implemented tools along with their names, descriptions, and examples of possible prompts.

Name	Description	Example Prompts
`query_aidefend`	Search the AIDEFEND AI security defense knowledge base. Use this to find defense strategies, techniques, and best practices for AI/ML security threats like prompt injection, model poisoning, data extraction, etc. Returns relevant defense tactics and implementation guidance.	"How to prevent prompt injection attacks in LLMs?" "What security measures protect against adversarial attacks on AI models?" "Tell me about defense strategies for supply chain attacks in ML pipelines"
`get_technique_detail`	Get complete details for a specific AIDEFEND technique including all sub-techniques, implementation strategies with code examples, tool recommendations, and threat mappings. This is the primary tool for deep-diving into a specific defense technique.	"Show me details for AID-H-001" "What does technique AID-D-005 include?" "Explain AID-H-002.003 in detail"
`get_defenses_for_threat`	Find AIDEFEND defense techniques for a specific threat. Supports threat IDs from OWASP LLM Top 10 (e.g., LLM01), MITRE ATLAS (e.g., T0043), MAESTRO, or natural language threat keywords (e.g., "prompt injection"). Essential for threat-driven defense planning.	"How to defend against LLM01?" "What defenses protect against model inversion attacks?" "Show me defenses for MAESTRO-003"
`get_secure_code_snippet`	Extract executable secure code snippets from AIDEFEND implementation strategies. Search by technique ID or topic keyword to get copy-paste-ready code examples. Perfect for developers implementing specific security controls.	"Show me code examples for AID-H-001" "Give me JavaScript code for implementing output validation" "Show implementation examples for RAG context isolation in Python"
`get_quick_reference`	Generate a quick reference guide for a specific security topic. Provides an actionable checklist organized by priority (quick wins, must-haves, nice-to-haves). Perfect for fast decision-making and presentations.	"Give me a quick reference for prompt injection defenses" "Create a checklist for securing AI agent systems" "I need quick wins for model poisoning prevention"
`get_implementation_plan`	Get ranked recommendations for next defense techniques to implement based on heuristic scoring (threat importance, ease of implementation, phase weight, pillar weight). Use this to prioritize security investments. Note: This tool provides ONLY heuristic scores. You should use these scores to make final recommendations via your own reasoning.	"What should I implement first for AI security?" "We've deployed AID-H-005, AID-D-003, and AID-H-001, recommend next steps" "Prioritize defenses for our LLM chatbot, nothing implemented yet"
`generate_incident_playbook`	Generate a structured incident response playbook based on threat classification. Provides a timeline-based action plan following NIST incident response phases: 1. Immediate Actions (0-15 min): Assessment, team activation, evidence preservation 2. Investigation (15 min - 2 hours): Threat classification, scope analysis, IOC collection 3. Containment (2-8 hours): Isolation, defense deployment, attack vector blocking 4. Recovery (8+ hours): Security controls, service restoration, post-incident review USE THIS when: Responding to an active AI/ML security incident Planning incident response procedures Training IR teams on AI-specific threats Documenting security incident workflows	"We detected suspicious prompt injection attempts, need incident response plan" "Anomalous queries extracting information from our RAG system detected" "User bypassed our content filters with adversarial prompts, need response steps"