Agent Immune

Overview

Agent Immune is a Model Context Protocol server that brings AI agent security to your workflow. It detects prompt injection attacks, maintains semantic memory, scans outputs for threats, and hardens prompts against adversarial input. Once connected, Claude, Cursor, VS Code, and other MCP clients can invoke these security tools directly.

The server runs locally on your machine using the agent-immune Python package via uvx, with no authentication required. It communicates over stdio and works immediately upon installation. Version v0.2.2 is available under the Apache-2.0 license, with source code at denial-web/agent-immune on GitHub.

Agent Immune is listed in the AI & Agents category and integrates with Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed, and other MCP clients. If you maintain this server, claim your listing to earn the Verified badge and keep your information current.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "agent_immune": {
      "command": "uvx",
      "args": [
        "agent-immune"
      ]
    }
  }
}