A security layer for AI agents that runs through MCP instead of adding more tools. Exposes 10 security-focused operations including input scanning (C3 firewall with 53 prompt injection patterns), response integrity checks (C4 immune system), action blocking (C5 guard with 80+ destructive patterns), and self-evolving prompt management across six chromosome layers. Works over stdio for Claude Desktop, Cursor, Windsurf, and other IDEs, or as an HTTP server with an OpenAI-compatible gateway for n8n, Make, and backend integrations. Supports Claude, GPT-4, Gemini, Ollama, and Perplexity. Useful when you need protection from injection attacks and unsafe operations without modifying your agent's core logic or workflow.
claude mcp add --transport stdio gsepcore-gsep-mcp uvx gsep-mcp