Llama MCP Streamlit

This project is an interactive AI assistant built with Streamlit, NVIDIA NIM's API (LLaMa 3.3:70b)/Ollama, and Model Control Protocol (MCP). It provides a conversational interface where you can interact with an LLM to execute real-time external tools via MCP, retrieve data, and perform actions seamlessly.

The assistant supports:

• Custom model selection (NVIDIA NIM / Ollama)
• API configuration for different backends
• Tool integration via MCP to enhance usability and real-time data processing
• A user-friendly chat-based experience with Streamlit

📸 Screenshots

📁 Project Structure

llama_mcp_streamlit/
│── ui/
│   ├── sidebar.py       # UI components for Streamlit sidebar
│   ├── chat_ui.py       # Chat interface components
│── utils/
│   ├── agent.py         # Handles interaction with LLM and tools
│   ├── mcp_client.py    # MCP client for connecting to external tools
│   ├── mcp_server.py    # Configuration for MCP server selection
│── config.py            # Configuration settings
│── main.py              # Entry point for the Streamlit app
.env                      # Environment variables
Dockerfile                # Docker configuration
pyproject.toml            # Poetry dependency management

🔧 Environment Variables

Before running the project, configure the .env file with your API keys:

# Endpoint for the NVIDIA Integrate API
API_ENDPOINT=https://integrate.api.nvidia.com/v1
API_KEY=your_api_key_here

# Endpoint for the Ollama API
API_ENDPOINT=http://localhost:11434/v1/
API_KEY=ollama

🚀 Running the Project

Using Poetry

1. Install dependencies:
   hljs language-bash

   Copy

   poetry install
   

2. Run the Streamlit app:
   hljs language-bash

   Copy

   poetry run streamlit run llama_mcp_streamlit/main.py
   

Using Docker

1. Build the Docker image:
   hljs language-bash

   Copy

   docker build -t llama-mcp-assistant .
   

2. Run the container:
   hljs language-bash

   Copy

   docker compose up
   

🔄 Changing MCP Server Configuration

To modify which MCP server to use, update the utils/mcp_server.py file. You can use either NPX or Docker as the MCP server:

NPX Server

server_params = StdioServerParameters(
    command="npx",
    args=[
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/username/Desktop",
        "/path/to/other/allowed/dir"
    ],
    env=None,
)

Docker Server

server_params = StdioServerParameters(
    command="docker",
    args=[
        "run",
        "-i",
        "--rm",
        "--mount", "type=bind,src=/Users/username/Desktop,dst=/projects/Desktop",
        "--mount", "type=bind,src=/path/to/other/allowed/dir,dst=/projects/other/allowed/dir,ro",
        "--mount", "type=bind,src=/path/to/file.txt,dst=/projects/path/to/file.txt",
        "mcp/filesystem",
        "/projects"
    ],
    env=None,
)

Modify the server_params configuration as needed to fit your setup.

📌 Features

• Real-time tool execution via MCP
• LLM-powered chat interface
• Streamlit UI with interactive chat elements
• Support for multiple LLM backends (NVIDIA NIM & Ollama)
• Docker support for easy deployment

🛠 Dependencies

• Python 3.11+
• Streamlit
• OpenAI API (for NVIDIA NIM integration)
• MCP (Model Control Protocol)
• Poetry (for dependency management)
• Docker (optional, for containerized deployment)

📜 License

This project is licensed under the MIT License.

🤝 Contributing

Feel free to submit pull requests or report issues!

📬 Contact

For any questions, reach out via GitHub Issues.