Agent Immune

Agent Immune is an open source MCP (Model Context Protocol) server built to enhance the security of AI agents by providing adaptive threat intelligence tools. It detects prompt injection, scans outputs for exfiltration/PII/credentials, provides semantic memory for learning attacks, and hardens prompts to prevent misuse. The server exposes methods like input/output assessment, threat learning, and real-time metrics, with deep integration for AI assistants (e.g. Claude, Cursor, Windsurf) and compatibility with deterministic governance stacks. It is valuable for developers, platform operators, and researchers securing LLM agents in local or production contexts.