MLOps MCP

FastMCP 机器学习工作流服务器

一个全面的MCP服务器，为药物发现、分子分析和一般数据科学工作流程提供机器学习工具。

快速入门

# Install dependencies
pip install -r requirements.txt

# Run the ML workflow MCP server (includes MLflow integration)
python fastmcp_server.py

使用示例

自然语言示例

使用MLflow跟踪的机器学习工作流程：

• *使用MLflow追踪功能在该数据集上训练一个随机森林模型*
• *运行MLflow快速入门演示*
• *“显示我最近的MLflow运行记录”*
• *“从我的训练实验中找到最佳模型”*
• *加载我的最佳模型以供服务*
• *为我的分子数据创建一个相关性热图*
• *评估我的训练模型并展示性能指标*

训练一个模型

from ml_workflow_server.tools.model_training import ModelTrainingTool

tool = ModelTrainingTool()
result = tool.execute(
    data_path="data/chembl_p53.csv",
    target_column="standard_value",
    model_type="random_forest_regressor",
    model_params={"n_estimators": 100, "random_state": 42},
    test_size=0.2
)
print(f"Model saved to: {result['model_path']}")

创建可视化图表

from ml_workflow_server.tools.data_visualization import DataVisualizationTool

viz_tool = DataVisualizationTool()
result = viz_tool.execute(
    data_path="data/protein_ligand_binding.csv",
    plot_type="correlation_heatmap",
    columns=["binding_affinity", "molecular_weight", "logp"],
    output_path="output/correlation_analysis.png",
    title="Molecular Property Correlations"
)

模型评估

from ml_workflow_server.tools.model_evaluation import ModelEvaluationTool

evaluator = ModelEvaluationTool()
eval_result = evaluator.execute(
    model_path="models/rf_model.pkl",
    test_data_path="data/test_dataset.csv",
    target_column="standard_value", 
    task_type="regression",
    output_dir="evaluation_results/"
)

克劳德桌面配置

要在Claude Desktop中使用此MCP服务器，请将以下配置添加到您的Claude Desktop设置中：

{
  "mcpServers": {
    "ml-workflow": {
      "command": "python",
      "args": ["fastmcp_server.py"],
      "cwd": "/your_workspace/MLOps_MCP"
    }
  }
}

项目结构

mlops-mcp/
├── fastmcp_server.py              # Main MCP server entry point
├── requirements.txt               # Python dependencies
└── ml_workflow_server/            # Core ML workflow package
    ├── tools/                     # ML tool implementations
    │   ├── data_preprocessing.py  # Data cleaning and feature engineering
    │   ├── data_visualization.py  # Advanced plotting and visualization
    │   ├── model_training.py      # ML model training pipeline
    │   └── model_evaluation.py    # Model evaluation and metrics
    └── utils/                     # Shared utilities
        ├── file_utils.py          # File operations and path management
        ├── mlflow_config.py       # MLflow configuration and logging
        └── mlflow_management.py   # MLflow model management