DeepTeam

An open-source LLM red-teaming and safety framework that simulates adversarial attacks—jailbreaking, prompt injection, multi-turn exploitation—to uncover vulnerabilities in AI agents, chatbots, and RAG pipelines. Also ships production-ready guards (ToxicityGuard, PromptInjectionGuard, PrivacyGuard) for real-time agent protection.

Connects to: OpenAI, DeepEval, custom LLMs · · Apache-2.0 2,410★

Use it with an AI agent

Loadbay is an MCP server, so an agent can search the catalog and find this harness:

claude mcp add --transport http loadbay https://loadbay.xyz/api/mcp

• Source: https://github.com/confident-ai/deepteam
• This harness as JSON: /api/harnesses/deepteam
• Agent setup: /setup.md
• Browse all 700+ harnesses on Loadbay