AI AI Red Team Engineering notes, research references, and field notes on AI systems, responsible AI, governance, and compliance.
Connect

Subscribe or connect programmatically

Use RSS if you want update polling in a feed reader. Use MCP if you want an AI client to search, read, and reason over the published library through a structured interface.

Open RSS Feed View MCP Setup
RSS

Feed reader access

Use the public feed URLs below with any RSS or Atom reader. The main feed tracks all published updates. Topic feeds let you subscribe to one area only.

Main feed
https://www.ai-redteam.com/rss.xml
Alias
https://www.ai-redteam.com/feed.xml
Topic feed pattern
https://www.ai-redteam.com/topics/<topic-slug>/rss.xml
Setup: copy one of the feed URLs into your reader, add the topic-specific pattern if you only want one area, and let the client poll for updates after new items are published.
ai-red-teamingprompt-engineeringprompt-injectionagent-securitymodel-evaluationai-compliance
MCP

AI client access

The public MCP endpoint is read-only and intended for AI clients, agent tooling, and remote MCP integrations that can talk to an HTTP endpoint.

Endpoint
https://www.ai-redteam.com/mcp
Transport
HTTP JSON-RPC over POST
Mode
Public read-only
Setup: add a remote MCP server in your client, point it at the endpoint above, and select HTTP or Streamable HTTP if your client distinguishes between MCP transports.
Example remote server definition
{
  "name": "ai-redteam",
  "type": "http",
  "url": "https://www.ai-redteam.com/mcp"
}

Client configuration formats vary. If your MCP client expects different field names, keep the same endpoint and choose the client’s HTTP-based remote MCP option.

What RSS Exposes
  • Published updates across the full library
  • Topic-specific feeds for narrower subscriptions
  • Titles, dates, summaries, categories, and source references
What MCP Exposes
  • Profile, topic, latest, featured, and per-insight resources
  • Search, latest-item, topic-overview, and insight lookup tools
  • Prompt templates and argument completions for AI clients
OpenClaw

Connecting OpenClaw to MCP

There are two realistic paths today: native OpenClaw MCP configuration, which should be treated as experimental for remote HTTP endpoints, and a plugin-based workaround that is built specifically for HTTP MCP servers.

Experimental

Native OpenClaw MCP setup

Recent OpenClaw docs and issue examples show MCP configuration support, but current releases have also reported failures for HTTP MCP servers with logs indicating that only stdio servers are supported. Use this path if you want to try native configuration first.

Example config shape
{
  "mcpServers": {
    "ai-redteam": {
      "type": "http",
      "url": "https://www.ai-redteam.com/mcp"
    }
  }
}
  • Enable the OpenClaw MCP command surface if you manage config through chat commands.
  • Add the server definition in your OpenClaw config or through the MCP management command.
  • Restart the gateway and confirm whether the server is discovered.
  • If OpenClaw logs that only stdio MCP servers are supported, switch to the plugin workaround below.
Workaround

HTTP MCP plugin route

A community OpenClaw MCP integration plugin exists specifically to bridge remote HTTP MCP servers into OpenClaw. Use this path if you need a practical way to connect to the public endpoint now.

Plugin config sketch
{
  "plugins": {
    "entries": {
      "mcp-integration": {
        "enabled": true,
        "config": {
          "enabled": true,
          "servers": {
            "ai-redteam": {
              "enabled": true,
              "transport": "http",
              "url": "https://www.ai-redteam.com/mcp"
            }
          }
        }
      }
    }
  }
}
  • Install the plugin into your OpenClaw extensions/plugins area.
  • Point the plugin at the MCP endpoint above.
  • Restart OpenClaw and use the plugin’s unified MCP tool surface to inspect available tools and call them.

This route depends on a third-party plugin rather than native OpenClaw support.