Introduction

Broken Guardrail AI collects current risks and best practices in AI and agent security. You can read it as a regular docs site, or plug it into your own agent as a skill or an MCP server. Your agent will then fetch the pages it needs on its own and act as a virtual AI security engineer you can consult.

Agent skill

Run this command to install the site as a skill in your own agent.

npx skills add https://ai.brokenguardrail.com

This installs a skill named brokenguardrail-ai. Once a task touches AI or agent security, it prompts the agent to fetch only the pages it needs as Markdown.

MCP server

You can connect to the site as an MCP server instead of installing the skill. In Claude Code, add it with this command within the project.

claude mcp add --transport http --scope project brokenguardrail-ai https://ai.brokenguardrail.com/mcp

That writes a .mcp.json file at the root of your project. To write it yourself, use this.

.mcp.json

{
  "mcpServers": {
    "brokenguardrail-ai": {
      "type": "http",
      "url": "https://ai.brokenguardrail.com/mcp"
    }
  }
}

Once connected, the agent can read the published skill and pages as resources. For the full set of configuration options, see the Claude Code MCP documentation.

Getting Started

Risks

Best Practices

Agent skill

MCP server

​Agent skill

​MCP server

Agent skill

MCP server