AlphaGenome MCP Server

Instructions

Implementation Reference

• scripts/alphagenome_bridge.py:622-656 (handler)

  Core implementation of batch_modality_screen: maps modality to specific output types, processes each variant by calling predict_variant_effect, collects and returns results.

  def batch_modality_screen(client, params: Dict[str, Any]) -> Dict[str, Any]:
      """Screen variants across specific modalities."""
      variants_data = params.get('variants', [])
      modality = params.get('modality', 'expression')
  
      # Map modality names to OutputType enums
      modality_map = {
          'expression': [dna_client.OutputType.RNA_SEQ, dna_client.OutputType.CAGE],
          'splicing': [dna_client.OutputType.SPLICE_SITES],
          'tf_binding': [dna_client.OutputType.CHIP_TF],
          'chromatin': [dna_client.OutputType.DNASE, dna_client.OutputType.ATAC]
      }
      modalities = modality_map.get(modality, [dna_client.OutputType.RNA_SEQ, dna_client.OutputType.SPLICE_SITES])
  
      results = []
      for v in variants_data:
          # Create a copy with output_types
          variant_params = v.copy()
          variant_params['output_types'] = modalities
          try:
              result = predict_variant_effect(client, variant_params)
              results.append({
                  'variant': result['variant'],
                  'predictions': result['predictions'],
                  'impact': result['interpretation']['impact_level']
              })
          except Exception as e:
              print(f"Warning: Failed: {e}", file=sys.stderr)
              continue
  
      return {
          'modalities_tested': [modality],  # Return string representation
          'total_variants': len(results),
          'results': results
      }

• src/index.ts:258-263 (handler)

  MCP server tool handler: invokes AlphaGenomeClient.batchModalityScreen and formats result as MCP content response.

  case 'batch_modality_screen': {
    const result = await getClient().batchModalityScreen(args);
    return {
      content: [{ type: 'text', text: JSON.stringify(result, null, 2) }],
    };
  }

• src/tools.ts:635-659 (schema)

  Input schema defining variants array (with chromosome, position, ref, alt) and modality enum.

  inputSchema: {
    type: 'object',
    properties: {
      variants: {
        type: 'array',
        items: {
          type: 'object',
          properties: {
            chromosome: { type: 'string' },
            position: { type: 'number' },
            ref: { type: 'string' },
            alt: { type: 'string' },
          },
          required: ['chromosome', 'position', 'ref', 'alt'],
        },
        minItems: 1,
      },
      modality: {
        type: 'string',
        enum: ['expression', 'splicing', 'tf_binding', 'chromatin'],
        description: 'Regulatory modality to screen',
      },
    },
    required: ['variants', 'modality'],
  },

• src/tools.ts:626-660 (registration)

  Tool definition and registration: exports BATCH_MODALITY_SCREEN_TOOL with name, description, and schema for MCP tool listing.

  export const BATCH_MODALITY_SCREEN_TOOL: Tool = {
    name: 'batch_modality_screen',
    description: `Screen variants across specific regulatory modalities.
  
  Efficiently tests multiple variants for specific regulatory effects.
  
  Perfect for: targeted regulatory screens, modality-specific studies.
  
  Example: "Screen 20 variants for splicing effects"`,
    inputSchema: {
      type: 'object',
      properties: {
        variants: {
          type: 'array',
          items: {
            type: 'object',
            properties: {
              chromosome: { type: 'string' },
              position: { type: 'number' },
              ref: { type: 'string' },
              alt: { type: 'string' },
            },
            required: ['chromosome', 'position', 'ref', 'alt'],
          },
          minItems: 1,
        },
        modality: {
          type: 'string',
          enum: ['expression', 'splicing', 'tf_binding', 'chromatin'],
          description: 'Regulatory modality to screen',
        },
      },
      required: ['variants', 'modality'],
    },
  };

• src/alphagenome-client.ts:436-442 (helper)

  Bridge method in AlphaGenomeClient: calls Python bridge script with 'batch_modality_screen' action.

  async batchModalityScreen(params: any): Promise<any> {
    try {
      return await this.callPythonBridge('batch_modality_screen', params);
    } catch (error) {
      if (error instanceof ApiError) throw error;
      throw new ApiError(`Batch modality screen failed: ${error}`, 500);
    }