Agent Mesh
Go a2aa2a-protocolagent-meshagentic-aiinferencemcpmeshmesh-networknetwork-securitynetworking-aip2p-network
Agent Mesh Chaos Agent
This directory contains a complete, production-grade Python AI Agent built with LangChain and the MCP Python SDK.
This provides a realistic workload for the Agent Mesh scale testing experiment. Rather than a synthetic benchmark or a hardcoded Go loop, this is a true autonomous agent capable of reasoning, exploring tools, and adversarial "Chaos Monkey" testing.
How it works
- It connects to the
agentmesh-nodeMCP endpoint via Streamable HTTP (oragentmesh-boxproxy in the microVM). - It dynamically converts all discovered MCP tools into LangChain
Toolobjects with JSON schemas. - It initializes a LangChain
create_agenttool-calling agent using the Agent Mesh's OpenAI-compatible inference endpoint (/v1/chat/completions). - It is given an adversarial prompt to explore all tools, fuzz them with extreme inputs, and find bugs.
- The LangChain framework automatically handles the multi-turn reasoning and tool-calling loop until the agent achieves its goal or hits the max iteration limit.
Running Locally
First, ensure you have Python installed, then install the dependencies:
pip install -r requirements.txt
Run the agent against a local agentmesh-node:
python agent.py --auth "Bearer <your-agentmesh-node-token>"
Customizing the Chaos
You can pass a custom prompt to instruct the agent to test specific behaviors:
python agent.py \
--prompt "Focus exclusively on the 'get_mesh_info' tool. Call it 5 times in rapid succession to test rate limiting." \
--auth "Bearer <token>"