Speech MCP Server

Instructions

Implementation Reference

• src/index.ts:309-317 (handler)

  MCP server request handler for the 'list_voices' tool. Invokes TTSClient.listVoices() and returns formatted list of voices.

  case "list_voices": {
    const voices = await ttsClient.listVoices();
    return {
      content: [{ 
        type: "text", 
        text: `Available voices:\n${voices.join('\n')}` 
      }],
    };
  }

• src/index.ts:218-228 (helper)

  Core logic in TTSClient class that lists and filters high-quality voices from the KokoroTTS instance after ensuring initialization.

  async listVoices(): Promise<KokoroVoice[]> {
    await this.waitForInit();
    if (!this.ttsInstance) {
      throw new Error("TTS model not initialized");
    }
    // @ts-ignore-line
    const allVoices = this.ttsInstance.voices as unknown as {[voice: string]: {overallGrade: string; gender: string}};
    const goodVoices = Object.keys(allVoices)
      .filter((voiceName) => ['A+', 'A', 'A-', 'B+', 'B', 'B-', 'C+'].includes(allVoices[voiceName].overallGrade))
    return goodVoices as unknown as KokoroVoice[];
  }

• src/index.ts:88-96 (schema)

  Tool schema definition including name, description, and empty input schema (no parameters required).

  const listVoicesTool: Tool = {
    name: "list_voices",
    description: "List all available voices for text-to-speech",
    inputSchema: {
      type: "object",
      properties: {},
      required: [],
    },
  };

• src/index.ts:355-360 (registration)

  Registration of listVoicesTool in the array returned by ListToolsRequest handler, making it discoverable by MCP clients.

  tools: [
    textToSpeechTool,
    textToSpeechWithOptionsTool,
    listVoicesTool,
    getModelStatusTool,
  ],