The Agent Prompt Injection Firewall MCP is an AI agent prompt injection firewall. It performs pattern-based and heuristic firewall scans on prompts, RAG documents, and tool arguments at runtime. The repository presents it as an OWASP LLM Top 10 #1 runtime guard.
🛠️ Key Features
- Pattern-based prompt firewall scanning
- Heuristic scanning of prompts and RAG documents
- Scanning of tool arguments
🚀 Use Cases
- Runtime protection against prompt injection attempts
- Securing RAG workflows from malicious document content
- Hardening agent tool usage by validating tool arguments
⚡ Developer Benefits
- MIT-licensed implementation
- Distributed as a PyPI package (
agent_prompt_injection_firewall_mcp)
⚠️ Limitations
- Description provides no details on rule sets, configuration, or supported MCP interfaces beyond scanning.
