OpenXAI MCP Server

Instructions

Implementation Reference

• index.js:304-405 (handler)

  The primary handler function implementing the list_datasets tool logic. It defines datasets by category and filters/returns them based on the input category parameter (default 'all'), formatting as MCP content response.

  async listDatasets(category) {
    const datasets = {
      synthetic: [
        {
          name: 'synthetic_classification',
          description: 'Synthetic classification dataset with ground truth explanations',
          task: 'classification',
          features: 'Customizable number of features',
          samples: 'Customizable number of samples'
        },
        {
          name: 'synthetic_regression',
          description: 'Synthetic regression dataset with ground truth explanations',
          task: 'regression',
          features: 'Customizable number of features',
          samples: 'Customizable number of samples'
        }
      ],
      'real-world': [
        {
          name: 'german',
          description: 'German Credit dataset - Binary classification for credit approval',
          task: 'classification',
          features: 20,
          samples: 1000,
          classes: 2
        },
        {
          name: 'compas',
          description: 'COMPAS Recidivism dataset - Binary classification for recidivism prediction',
          task: 'classification',
          features: 11,
          samples: 6172,
          classes: 2
        },
        {
          name: 'adult',
          description: 'Adult Income dataset - Binary classification for income prediction',
          task: 'classification',
          features: 14,
          samples: 48842,
          classes: 2
        },
        {
          name: 'folktable',
          description: 'ACS Folktables dataset - Various prediction tasks',
          task: 'classification',
          features: 'Variable',
          samples: 'Variable',
          classes: 'Variable'
        }
      ],
      tabular: [
        'german', 'compas', 'adult', 'folktable', 'synthetic_classification', 'synthetic_regression'
      ],
      image: [
        {
          name: 'mnist',
          description: 'MNIST handwritten digits dataset',
          task: 'classification',
          features: '28x28 grayscale images',
          samples: 70000,
          classes: 10
        },
        {
          name: 'cifar10',
          description: 'CIFAR-10 object recognition dataset',
          task: 'classification',
          features: '32x32 color images',
          samples: 60000,
          classes: 10
        }
      ],
      text: [
        {
          name: 'imdb',
          description: 'IMDB Movie Review sentiment classification',
          task: 'classification',
          features: 'Text sequences',
          samples: 50000,
          classes: 2
        }
      ]
    };
  
    let result = [];
    if (category === 'all') {
      result = Object.values(datasets).flat();
    } else {
      result = datasets[category] || [];
    }
  
    return {
      content: [
        {
          type: 'text',
          text: `Available OpenXAI datasets (${category}):\n\n` +
                JSON.stringify(result, null, 2)
        }
      ]
    };
  }

• index.js:41-51 (schema)

  The input schema for the list_datasets tool, specifying an optional 'category' string parameter with allowed enum values for filtering datasets.

  inputSchema: {
    type: 'object',
    properties: {
      category: {
        type: 'string',
        description: 'Filter by dataset category (synthetic, real-world, tabular, image, text)',
        enum: ['synthetic', 'real-world', 'tabular', 'image', 'text', 'all']
      }
    },
    required: []
  }

• index.js:38-52 (registration)

  Registration of the list_datasets tool in the tools list returned by ListToolsRequestSchema handler.

  {
    name: 'list_datasets',
    description: 'List available datasets in OpenXAI framework',
    inputSchema: {
      type: 'object',
      properties: {
        category: {
          type: 'string',
          description: 'Filter by dataset category (synthetic, real-world, tabular, image, text)',
          enum: ['synthetic', 'real-world', 'tabular', 'image', 'text', 'all']
        }
      },
      required: []
    }
  },

• index.js:255-257 (registration)

  Dispatch/registration of the list_datasets handler in the CallToolRequestSchema switch statement.

  case 'list_datasets':
    return await this.listDatasets(args.category || 'all');