AlphaFold MCP Server

Instructions

Implementation Reference

• src/index.ts:925-994 (handler)

  The main execution handler for the 'get_confidence_scores' tool. Validates input, fetches AlphaFold prediction data, generates mock per-residue confidence scores (pLDDT-like), applies optional threshold filter, computes summary statistics, and returns JSON-formatted results.

  private async handleGetConfidenceScores(args: any) {
    if (!isValidConfidenceArgs(args)) {
      throw new McpError(ErrorCode.InvalidParams, 'Invalid confidence score arguments');
    }
  
    try {
      const response = await this.apiClient.get(`/prediction/${args.uniprotId}`);
      const structures = response.data;
  
      if (!structures || structures.length === 0) {
        return {
          content: [
            {
              type: 'text',
              text: `No structure available for ${args.uniprotId}`,
            },
          ],
        };
      }
  
      const structure = structures[0];
  
      // Mock confidence data based on sequence length
      const confidenceData: ConfidenceData[] = [];
      const sequenceLength = structure.uniprotSequence.length;
  
      for (let i = 1; i <= sequenceLength; i++) {
        // Generate mock confidence scores (in real implementation, this would come from the API)
        const score = Math.random() * 100;
        const category = score >= 90 ? 'very-high' :
                        score >= 70 ? 'confident' :
                        score >= 50 ? 'low' : 'very-low';
  
        if (!args.threshold || score >= args.threshold) {
          confidenceData.push({
            residueNumber: i,
            confidenceScore: score,
            confidenceCategory: category as any,
          });
        }
      }
  
      return {
        content: [
          {
            type: 'text',
            text: JSON.stringify({
              uniprotId: args.uniprotId,
              confidenceScores: confidenceData,
              summary: {
                totalResidues: sequenceLength,
                filteredResidues: confidenceData.length,
                averageConfidence: confidenceData.reduce((sum, c) => sum + c.confidenceScore, 0) / confidenceData.length,
              },
            }, null, 2),
          },
        ],
      };
    } catch (error) {
      return {
        content: [
          {
            type: 'text',
            text: `Error fetching confidence scores: ${error instanceof Error ? error.message : 'Unknown error'}`,
          },
        ],
        isError: true,
      };
    }
  }

• src/index.ts:593-594 (registration)

  Dispatch registration in the CallToolRequestSchema handler switch statement, routing calls to 'get_confidence_scores' to the handleGetConfidenceScores method.

  case 'get_confidence_scores':
    return this.handleGetConfidenceScores(args);

• src/index.ts:421-431 (schema)

  Tool metadata registration including name, description, and input schema definition (uniprotId required string, optional threshold number 0-100) in the ListToolsRequestSchema response.

    name: 'get_confidence_scores',
    description: 'Get per-residue confidence scores for a structure prediction',
    inputSchema: {
      type: 'object',
      properties: {
        uniprotId: { type: 'string', description: 'UniProt accession' },
        threshold: { type: 'number', description: 'Confidence threshold (0-100, optional)', minimum: 0, maximum: 100 },
      },
      required: ['uniprotId'],
    },
  },

• src/index.ts:122-132 (helper)

  Type guard and validation helper specifically for 'get_confidence_scores' tool arguments, ensuring uniprotId is a non-empty string and threshold is optional number in [0,100].

  const isValidConfidenceArgs = (
    args: any
  ): args is { uniprotId: string; threshold?: number } => {
    return (
      typeof args === 'object' &&
      args !== null &&
      typeof args.uniprotId === 'string' &&
      args.uniprotId.length > 0 &&
      (args.threshold === undefined || (typeof args.threshold === 'number' && args.threshold >= 0 && args.threshold <= 100))
    );
  };