MCP Agent Bot Conversation

This project demonstrates a multi-agent conversation system where two language model agents (BotA and BotB) interact with each other, orchestrated by a driver script. The agents utilize a local LLM served by Ollama, and the communication framework potentially leverages the Model Context Protocol (MCP), although the current implementation uses direct interactions defined within the bot scripts.

Features

• Multi-Agent Interaction: Simulates a conversation between two distinct AI bots (BotA and BotB).
• Local LLM Powered: Uses a locally hosted language model via Ollama (configured for phi3 by default).
• Dockerized Ollama: Includes a Docker setup to easily run the required Ollama server and automatically pull the specified model.
• MCP Framework: Built using the mcp Python SDK for the bot servers.
• Conversation Logging: Records the full conversation history with timestamps in src/conversation_log.txt.

Technology Stack

• Python 3.12+
• Ollama (running phi3 model by default)
• Model Context Protocol (MCP) Python SDK
• FastAPI & Uvicorn (for bot servers)
• OpenAI Python Client (configured for Ollama)
• Docker

Project Structure

mcp-agent/
├── Dockerfile                  # Defines the Docker image for running Ollama
├── README.md                   # This file
├── docker-entrypoint-ollama.sh # Helper script run by Dockerfile to start Ollama & pull model
└── src/
    ├── .env                    # Optional file for environment variables (e.g., OPENAI_API_KEY, if used)
    ├── bot_a.py                # Runs the MCP server for Bot A (listens on port 8001)
    ├── bot_b.py                # Runs the MCP server for Bot B (listens on port 8002)
    ├── converse.py             # Driver script orchestrating the conversation
    ├── requirements.txt        # Python dependencies
    ├── conversation_log.txt    # Log file for conversations (created on run)
    └── venv/                   # Optional: Python virtual environment directory

Setup

1. Clone the Repository:

   git clone https://github.com/vishalmakwana111/mcp-agent.git
   cd mcp-agent

2. Install Docker: Ensure you have Docker installed and running. (Install Docker)
3. Build the Ollama Docker Image: This image includes Ollama and automatically pulls the required model (phi3 by default) when started.

   docker build -t ollama-server-with-model .

4. Set up Python Environment: It's recommended to use a virtual environment for the Python scripts.

   # Navigate to the source directory
   cd src
   
   # Create a virtual environment (if you haven't already)
   python -m venv venv
   
   # Activate the virtual environment
   # On Windows (Powershell):
   .\venv\Scripts\Activate.ps1
   # On macOS/Linux:
   # source venv/bin/activate
   
   # Install Python dependencies
   pip install -r requirements.txt

Running the Application

1. Start the Ollama Docker Container: Run the Ollama server in the background. This command also maps the necessary port.

   # Run this from the project root directory (mcp-agent/)
   docker run -d --rm --name ollama -p 11434:11434 ollama-server-with-model

   • You can check the logs to see the model being pulled: docker logs -f ollama (Press Ctrl+C to stop following). Wait for the "Model pull complete" message before proceeding.

2. Start Bot A:

   • Open a new terminal.
   • Navigate to the src directory: cd path/to/mcp-agent/src
   • Activate the virtual environment (e.g., .\venv\Scripts\Activate.ps1).
   • Run the bot: python bot_a.py

3. Start Bot B:

   • Open another new terminal.
   • Navigate to the src directory: cd path/to/mcp-agent/src
   • Activate the virtual environment.
   • Run the bot: python bot_b.py

4. Run the Conversation Driver:

   • Open yet another new terminal.
   • Navigate to the src directory: cd path/to/mcp-agent/src
   • Activate the virtual environment.
   • Run the driver script: python converse.py

Configuration

• LLM Model: The model used by the bots (phi3) is configured in bot_a.py, bot_b.py, and docker-entrypoint-ollama.sh. Change the model name in all three places if you want to use a different Ollama model.
• Ollama URL: The bots connect to http://localhost:11434/v1. This is configured in bot_a.py and bot_b.py.
• Bot Prompts: The system prompts defining the bots' personas are set within bot_a.py and bot_b.py.
• Log File: Conversation output is saved to src/conversation_log.txt.
• Conversation Length/Delay: The number of turns and the delay between turns are set in the for loop and asyncio.sleep call within src/converse.py.