Void MCP Server

Void MCP Server

This project provides a Model Context Protocol (MCP) server for storing and retrieving LLM context data. It ships with run_local.py for running a lightweight Hugging Face model and app.py for a simple Streamlit UI.

Quick Start

1. Install prerequisites

   • Node.js and npm
   • Python 3.11
   • PostgreSQL (ensure the service is running)

2. Install dependencies

   npm install          # Node packages
   pip install -r requirements.txt
   

3. Configure the database

   Create a .env file in the project root and set at least the following variables:

   DATABASE_HOST=localhost
   DATABASE_PORT=5432
   DATABASE_NAME=llm_context_db
   DATABASE_USER=your_db_user
   DATABASE_PASSWORD=your_db_password  # must be a string
   PORT=3000
   JWT_SECRET=change-me-in-production-min-32-chars
   

   If DATABASE_PASSWORD is missing you will see the error: SASL: SCRAM-SERVER-FIRST-MESSAGE: client password must be a string.

4. Build the server

   npm run build
   

5. Run everything

   python start.py
   

   The MCP API will be available at http://localhost:3000 and the local model server at http://localhost:8000. Press Ctrl+C to stop them.

6. Optional: Launch the Streamlit interface

   streamlit run app.py
   

Development Tips

• Run tests with pytest and npm test.

• Start only the local model using python run_local.py --model sshleifer/tiny-gpt2 --port 8000.

• Format and lint with pre-commit run --all-files.

• setup.sh provides an interactive database setup if you need it.

• Install dependencies

  • Node: npm install
  • Python: pip install -r requirements.txt

• Build: npm run build

• Tests: npm test for Node, pytest for Python

• Run local model: python run_local.py --model sshleifer/tiny-gpt2 --port 8000

• Streamlit UI: streamlit run app.py

One-Click Start

To launch both the MCP server and a local model server in one step, run:

python start.py

The start.py script is cross-platform and installs any missing Node and Python dependencies, builds the TypeScript server if necessary, and then starts the MCP server alongside the local model server. On Unix-like systems you can also use ./start.sh.

The script installs missing Node dependencies, builds the server if needed, and then starts the MCP service alongside the local model server. Connection details are printed to the terminal for easy integration with IDEs and tools.

IDE Integration

To connect your IDE's context features to the MCP server, provide a configuration file like ide_config.json pointing to start.py in this repository.

Replace /path/to/Void-MCP-Server/start.py with the absolute path on your system:

{
  "context_servers": {
    "my-mcp-server": {
      "command": {
        "path": "python",
        "args": [
          "/path/to/Void-MCP-Server/start.py"
        ],
        "env": {}
      },
      "settings": {}
    }
  }
}

Run python start.py --help for available options such as --model, --port, and --model-port to customize the servers.