# AI Red Teaming MCP Server 🎯⚡

> **Jailbreak LLMs. Bypass guardrails. Find the holes before attackers do.**

AI red teaming toolkit exposed as a Model Context Protocol (MCP) server. Connect any MCP client and test AI systems for safety vulnerabilities.

## What It Does

| Tool | What It Tests |
|------|---------------|
| `redteam_jailbreak` | LLM jailbreak resistance (instruction, roleplay, encoding, translation, composite) |
| `redteam_prompt_injection` | Direct, indirect, system prompt leakage, data exfiltration |
| `redteam_guardrail_bypass` | Encoding, obfuscation, context manipulation, token splitting |
| `redteam_model_scan` | Bias, backdoors, data leakage, hallucination patterns |
| `redteam_challenges` | CTF-style AI safety challenges (LLM CTF, HarmBench) |

## Install

```bash
pip install mcp-server-ai-redteam
# OR
git clone https://github.com/manteclaw/mcp-server-ai-redteam.git
cd mcp-server-ai-redteam
pip install -r requirements.txt
```

## Connect

### Claude Desktop

```json
{
  "mcpServers": {
    "ai-redteam": {
      "command": "python",
      "args": ["/path/to/server.py"]
    }
  }
}
```

### OpenClaw

```bash
openclaw skill add projects/mcp-marketplace/mcp-server-ai-redteam
```

## Usage Examples

> "Test this model for jailbreak vulnerabilities using roleplay payloads"

> "Fuzz this prompt for injection vectors"

> "Scan this endpoint for bias and backdoor patterns"

## Pricing

| Tier | Price | Limits |
|------|-------|--------|
| Community | Free | 20 calls/day |
| Pro | $39/month | Unlimited + custom payloads |
| Enterprise | $299/month | Unlimited + custom models + SLA |

## License

MIT

---

*Built by Manteclaw | AI safety through adversarial testing*