AgentTorch MCP Server

Imagine if you could turn an LLM into a simulator

Interface for turning AgentTorch into an MCP server - build, evaluate and analyze simulations.

Features

• Dark Mode UI: Easy on the eyes with a modern dark interface
• Claude-like Chat Interface: Interact naturally with the simulation system
• Real-time Visualization: See simulation progress and population dynamics
• LLM-powered Analysis: Get intelligent insights about simulation behavior
• Sample Prompts: Quick-start with pre-written questions and scenarios

Setup

1. Make sure you have the required Python packages:

   pip install -r requirements.txt
   

2. Ensure you have set the ANTHROPIC_API_KEY environment variable:

   export ANTHROPIC_API_KEY=your_api_key_here
   

3. Verify that the data directory exists at the correct location:

   services/data/18x25/
   

Running the Server

Start the server with:

python server.py

Then access the interface at http://localhost:8000

How to Use

1. Ask a Question: Type a question in the input box or select a sample prompt
2. Run Simulation: Click "Run Simulation & Analyze" to start the process
3. Watch Simulation: View real-time logs and progress updates
4. See Results: When complete, the population chart will be displayed
5. Get Analysis: The LLM will automatically analyze the results based on your question

Sample Prompts

The interface includes several sample prompts you can try:

• What happens to prey population when predators increase?
• How does the availability of food affect the predator-prey dynamics?
• What emergent behaviors appear in this ecosystem?
• Analyze the oscillations in population levels over time
• What would happen if the nutritional value of grass was doubled?

Project Structure

├── server.py           # Main FastAPI server
├── requirements.txt    # Dependencies
├── static/             # Static CSS files
│   └── styles.css      # Dark mode styling
├── templates/          # HTML templates
│   └── index.html      # Main UI with chat interface
├── services/           # Service layer
│   ├── simulation.py   # Simulation service using AgentTorch
│   ├── llm.py          # LLM service using Claude API
│   └── data/           # Simulation data files
│       └── 18x25/      # Grid size specific data files

Technical Notes

• The simulation uses AgentTorch framework and the provided config.yaml
• WebSockets enable real-time updates during simulation
• The UI is designed to work well on both desktop and mobile devices
• LLM analysis is powered by the Claude API