mloop-mcp

MCP (Model Context Protocol) server for MLoop CLI.

Enables AI clients (Claude, Cursor, ironhive-cli) to perform MLOps tasks using MLoop's AutoML capabilities.

Installation

npm install -g @iyulab/mloop-mcp

Or run directly with npx:

npx @iyulab/mloop-mcp

Prerequisites

• Node.js 18+
• MLoop CLI installed and available in PATH

  dotnet tool install -g mloop

Custom MLoop Path

If MLoop is not in PATH, set the MLOOP_PATH environment variable:

# Windows
set MLOOP_PATH=D:\lib\mloop.exe

# Linux/macOS
export MLOOP_PATH=/usr/local/bin/mloop

Configuration

For Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "mloop": {
      "command": "npx",
      "args": ["@iyulab/mloop-mcp"]
    }
  }
}

For ironhive-cli

Add to .ironhive/plugins.yaml:

plugins:
  mloop:
    transport: stdio
    command: npx
    args:
      - "@iyulab/mloop-mcp"

Available Tools

| Tool | Description | |------|-------------| | mloop_train | Train ML models using AutoML | | mloop_predict | Run predictions with trained models | | mloop_list | List experiments and their metrics | | mloop_promote | Promote an experiment to production | | mloop_info | Analyze and profile datasets | | mloop_status | Show project status | | mloop_compare | Compare multiple experiments | | mloop_evaluate | Evaluate model performance | | mloop_serve | Start REST API server |

Tool Examples

Train a model

{
  "tool": "mloop_train",
  "arguments": {
    "projectPath": "/path/to/project",
    "dataFile": "datasets/train.csv",
    "label": "target",
    "task": "binary-classification",
    "time": 60
  }
}

Run predictions

{
  "tool": "mloop_predict",
  "arguments": {
    "projectPath": "/path/to/project",
    "dataFile": "datasets/test.csv",
    "output": "predictions/output.csv"
  }
}

List experiments

{
  "tool": "mloop_list",
  "arguments": {
    "projectPath": "/path/to/project",
    "showAll": true
  }
}

Promote to production

{
  "tool": "mloop_promote",
  "arguments": {
    "projectPath": "/path/to/project",
    "experimentId": "exp-003",
    "force": true
  }
}

Analyze dataset

{
  "tool": "mloop_info",
  "arguments": {
    "dataFile": "datasets/train.csv",
    "projectPath": "/path/to/project"
  }
}

Development

# Install dependencies
npm install

# Build
npm run build

# Run locally
npm start

# Test with MCP Inspector
npx @modelcontextprotocol/inspector node build/index.js

Architecture

ironhive-cli ──[MCP/STDIO]──> mloop-mcp ──[subprocess]──> mloop CLI ──> ML.NET

• Pure Bridge Pattern: 1:1 CLI mapping, no business logic, stateless
• STDIO Transport: Simple process communication
• Subprocess Execution: Each tool call spawns mloop CLI

License

MIT