OpenXAI MCP Server

Instructions

Implementation Reference

• index.js:466-523 (handler)

  The handler function that implements the list_models tool logic. It defines available models, filters them by dataset_name and model_type, and returns a formatted list.

  async listModels(datasetName, modelType) {
    const models = {
      ann: {
        name: 'Artificial Neural Network',
        description: 'Multi-layer perceptron with configurable architecture',
        supported_datasets: ['german', 'compas', 'adult', 'folktable', 'mnist', 'cifar10'],
        task_types: ['classification', 'regression']
      },
      lr: {
        name: 'Logistic Regression',
        description: 'Linear model for classification with ground truth explanations',
        supported_datasets: ['german', 'compas', 'adult', 'folktable'],
        task_types: ['classification']
      },
      rf: {
        name: 'Random Forest',
        description: 'Ensemble of decision trees',
        supported_datasets: ['german', 'compas', 'adult', 'folktable'],
        task_types: ['classification', 'regression']
      },
      svm: {
        name: 'Support Vector Machine',
        description: 'Kernel-based classification model',
        supported_datasets: ['german', 'compas', 'adult', 'folktable'],
        task_types: ['classification']
      },
      xgb: {
        name: 'XGBoost',
        description: 'Gradient boosting framework',
        supported_datasets: ['german', 'compas', 'adult', 'folktable'],
        task_types: ['classification', 'regression']
      }
    };
  
    let result = [];
    if (modelType === 'all') {
      result = Object.entries(models).map(([key, value]) => ({
        type: key,
        ...value
      }));
    } else {
      result = models[modelType] ? [{ type: modelType, ...models[modelType] }] : [];
    }
  
    if (datasetName) {
      result = result.filter(model => model.supported_datasets.includes(datasetName));
    }
  
    return {
      content: [
        {
          type: 'text',
          text: `Available OpenXAI models${datasetName ? ` for dataset '${datasetName}'` : ''}:\n\n` +
                JSON.stringify(result, null, 2)
        }
      ]
    };
  }

• index.js:75-89 (schema)

  Input schema definition for the list_models tool, specifying optional parameters dataset_name and model_type with enums.

  inputSchema: {
    type: 'object',
    properties: {
      dataset_name: {
        type: 'string',
        description: 'Filter models by dataset they were trained on'
      },
      model_type: {
        type: 'string',
        description: 'Filter by model type (ann, lr, rf, etc.)',
        enum: ['ann', 'lr', 'rf', 'svm', 'xgb', 'all']
      }
    },
    required: []
  }

• index.js:72-90 (registration)

  Registration of the list_models tool in the MCP server's tools list, including name, description, and schema.

  {
    name: 'list_models',
    description: 'List available pre-trained models in OpenXAI',
    inputSchema: {
      type: 'object',
      properties: {
        dataset_name: {
          type: 'string',
          description: 'Filter models by dataset they were trained on'
        },
        model_type: {
          type: 'string',
          description: 'Filter by model type (ann, lr, rf, etc.)',
          enum: ['ann', 'lr', 'rf', 'svm', 'xgb', 'all']
        }
      },
      required: []
    }
  },

• index.js:261-263 (registration)

  Dispatch case in the CallToolRequestSchema handler that routes list_models calls to the listModels method.

  case 'list_models':
    return await this.listModels(args.dataset_name, args.model_type || 'all');