ml-training-init-mcp 0.1.0

MCP server for ML training script initialization with file locking

ML Training Init MCP Server

A Model Context Protocol (MCP) server that implements a sequential thinking pattern for ML training script generation with strict file locking constraints.

Key Feature: Single File Lock Constraint

IMPORTANT: This server enforces a strict single-file constraint:

• Once a training file is created, the server locks to ONLY that file
• ALL subsequent operations must work with this single locked file
• No new files can be created after initialization
• This prevents agents from creating unnecessary files when stuck

Installation

pip install -r requirements.txt

Running the Server

python -m src.server

Usage with Claude Desktop

Add to your Claude Desktop configuration:

{
  "mcpServers": {
    "ml-training-init": {
      "command": "python",
      "args": ["-m", "src.server"],
      "cwd": "/path/to/ml-training-init-mcp"
    }
  }
}

Available Tools

1. initialize_training_file

• Creates ONE training file and permanently locks to it
• Takes: file_name, use_case, libraries, reference_code
• Returns: file_path, checkpoint_id, locked_file status

2. enhance_code

• Modifies ONLY the locked file
• Cannot create new files
• Takes: enhancement_type, instructions
• Returns: status, checkpoint_id

3. get_current_file

• Returns content of the locked file
• Use this instead of trying to create new files
• Returns: file_path, content, checkpoint_id

Testing

Run the test example:

python test_example.py

This will show you how to interact with the MCP server through Claude.