Off-the-Shelf Agents

AgentSuite-Red is designed to stress your existing agent stack without forcing you to rewrite it. If your agent is built on any of the frameworks listed below, you can connect it to AgentSuite-Red as-is — the only thing we need from you is an HTTP endpoint and an MCP-capable agent (see Connect Your Agent).

Supported agent frameworks and runtimes

The following frameworks and agent products work with AgentSuite-Red out of the box. "Out of the box" means: your agent runs on your side using its native SDK, AgentSuite-Red calls your HTTP endpoint, and your agent consumes the ForgingGround MCP server like any other MCP server.

Foundation-model agent SDKs

Framework	Description
OpenAI Agents SDK	Official OpenAI Python SDK for agent applications. Native MCP support, structured outputs, function calling, tracing.
OpenAI Codex	OpenAI's code-focused agent runtime, suitable for software-engineering domain red-teaming.
Claude Agent SDK	Anthropic's official agent SDK with computer-use, tool-use, and vision support.
Claude Code	Anthropic's official CLI agent for software-engineering tasks. Connect via Claude Code's MCP server config to point at the ForgingGround endpoint.
Claude Cowork	Anthropic's collaborative-agent runtime — connect by registering ForgingGround as an external MCP.
Google ADK	Google's Agent Development Kit for Gemini models. Multimodal, grounding, extensions.
AWS Bedrock Agents	Managed agent runtime on AWS Bedrock. Wire ForgingGround in as an MCP-backed action group via the Bedrock agent's tool config.
Microsoft Agent Studio	Microsoft's agent runtime (formerly Copilot Studio agent builder). Connect via custom MCP connector.

Orchestration frameworks

Framework	Description
LangChain + LangGraph	Most popular orchestration framework for LLM apps. Use the LangChain MCP adapter to attach the ForgingGround MCP server.
OpenClaw	Open-source agent CLI with first-class MCP support — connect by adding ForgingGround as an MCP server in the OpenClaw config.
NanoClaw	Lightweight sibling of OpenClaw aimed at smaller-footprint deployments. Same MCP wiring.
Hermes	Tool-using agent runtime built around the Hermes function-calling models. Register ForgingGround as an MCP tool server and Hermes can drive it like any other tool source.

Coding agents and IDEs

Framework	Description
Cursor	Cursor's agent uses MCP natively — add ForgingGround in ~/.cursor/mcp.json.
Claude Code	See above; works as a coding agent under software-development domains.
OpenAI Codex	See above.

And many more

If your agent speaks MCP and you can expose it behind an HTTP endpoint that matches our request/response contract, AgentSuite-Red can drive it — regardless of the underlying framework. That includes:

• Local agents built on top of LiteLLM, Together AI, Groq, vLLM, or Ollama
• Custom in-house agents written directly against provider APIs
• Agent runtimes from Cohere, Mistral, Fireworks, Databricks, and other providers
• Any framework that exposes a function-calling/tool-use surface compatible with MCP

The Connect Your Agent page walks through the minimal contract; if your stack can satisfy it, you are supported.

Minimum requirements

Whichever framework you use, your agent must:

1. Be reachable over HTTP at an endpoint you control, accepting query + optional session_id and returning a string response.
2. Support MCP (Streamable HTTP) so it can call the ForgingGround MCP server at https://agentsuite-red.virtueai.cc/forgingground/mcp with an X-API-Key header.
3. Honor session_id across calls so AgentSuite-Red can run multi-turn red-teaming scenarios against your agent.

What's next

• Pick your framework above and follow Connect Your Agent to wire it up.
• For coverage details on what AgentSuite-Red will exercise, see Domains and Environments.