text_to_speech

Implementation Reference

• src/index.js:66-112 (registration)

  The `text_to_speech` tool is registered on the MCP server via `server.tool(...)`. It accepts text, output_path, optional voice, and optional model parameters. The handler calls the OpenAI-compatible TTS API, writes the audio buffer to disk, and returns a confirmation message.

  server.tool(
    "text_to_speech",
    {
      text: z.string().describe("Text to convert to speech"),
      output_path: z
        .string()
        .describe("Absolute path to write the output audio file (e.g. /tmp/output.mp3)"),
      voice: z
        .string()
        .optional()
        .describe(`Voice ID to use. Defaults to ${TTS_VOICE}`),
      model: z
        .string()
        .optional()
        .describe(`TTS model ID to use. Defaults to ${TTS_MODEL}`),
    },
    async ({ text, output_path, voice, model }) => {
      const dir = path.dirname(output_path);
      if (!fs.existsSync(dir)) {
        return {
          content: [{ type: "text", text: `Error: output directory does not exist: ${dir}` }],
          isError: true,
        };
      }
  
      let buffer;
      try {
        const response = await client.audio.speech.create({
          model: model || TTS_MODEL,
          voice: voice || TTS_VOICE,
          input: text,
        });
        buffer = Buffer.from(await response.arrayBuffer());
      } catch (err) {
        return {
          content: [{ type: "text", text: `Error: TTS failed: ${err.message}` }],
          isError: true,
        };
      }
  
      fs.writeFileSync(output_path, buffer);
  
      return {
        content: [{ type: "text", text: `Audio saved to ${output_path}` }],
      };
    }
  );

• src/index.js:68-81 (schema)

  Input schema for the `text_to_speech` tool, defined using Zod: `text` (required string), `output_path` (required string), `voice` (optional string, defaults to TTS_VOICE env var), and `model` (optional string, defaults to TTS_MODEL env var).

  {
    text: z.string().describe("Text to convert to speech"),
    output_path: z
      .string()
      .describe("Absolute path to write the output audio file (e.g. /tmp/output.mp3)"),
    voice: z
      .string()
      .optional()
      .describe(`Voice ID to use. Defaults to ${TTS_VOICE}`),
    model: z
      .string()
      .optional()
      .describe(`TTS model ID to use. Defaults to ${TTS_MODEL}`),
  },

• src/index.js:82-111 (handler)

  The handler function for `text_to_speech`. It validates the output directory exists, calls `client.audio.speech.create()` (OpenAI-compatible TTS API), buffers the response, writes it to the output file, and returns a success or error message.

  async ({ text, output_path, voice, model }) => {
    const dir = path.dirname(output_path);
    if (!fs.existsSync(dir)) {
      return {
        content: [{ type: "text", text: `Error: output directory does not exist: ${dir}` }],
        isError: true,
      };
    }
  
    let buffer;
    try {
      const response = await client.audio.speech.create({
        model: model || TTS_MODEL,
        voice: voice || TTS_VOICE,
        input: text,
      });
      buffer = Buffer.from(await response.arrayBuffer());
    } catch (err) {
      return {
        content: [{ type: "text", text: `Error: TTS failed: ${err.message}` }],
        isError: true,
      };
    }
  
    fs.writeFileSync(output_path, buffer);
  
    return {
      content: [{ type: "text", text: `Audio saved to ${output_path}` }],
    };
  }