Pollinations Multimodal MCP Server

Instructions

Implementation Reference

• src/services/audioService.js:126-208 (handler)

  The handler function that executes the sayText tool: generates verbatim text-to-speech audio using the Pollinations API, supports voice, format, instructions, and optional playback.

  async function sayText(params) {
      const {
          text,
          voice = "alloy",
          format = "mp3",
          voiceInstructions,
          audioPlayer,
          tempDir,
      } = params;
  
      if (!text || typeof text !== "string") {
          throw new Error("Text is required and must be a string");
      }
  
      // Prepare the query parameters
      const queryParams = {
          model: "openai-audio",
          voice,
          format,
      };
  
      // Prepare the prompt with the verbatim instruction
      let finalPrompt = `Say verbatim: ${text}`;
  
      // Add voice instructions if provided
      if (voiceInstructions) {
          finalPrompt = `${voiceInstructions}\n\n${finalPrompt}`;
      }
  
      // Build the URL using the utility function
      const url = buildUrl(
          AUDIO_API_BASE_URL,
          encodeURIComponent(finalPrompt),
          queryParams,
      );
  
      try {
          // Fetch the audio from the URL
          const response = await fetch(url);
  
          if (!response.ok) {
              throw new Error(
                  `Failed to generate speech: ${response.statusText}`,
              );
          }
  
          // Get the audio data as an ArrayBuffer
          const audioBuffer = await response.arrayBuffer();
  
          // Convert the ArrayBuffer to a base64 string
          const base64Data = Buffer.from(audioBuffer).toString("base64");
  
          // Determine the mime type from the format
          const mimeType = `audio/${format === "mp3" ? "mpeg" : format}`;
  
          // Play the audio if an audio player is provided
          if (audioPlayer) {
              const tempDirPath = tempDir || os.tmpdir();
              await playAudio(
                  base64Data,
                  mimeType,
                  "say_text",
                  audioPlayer,
                  tempDirPath,
              );
          }
  
          // Return the response in MCP format
          return createMCPResponse([
              {
                  type: "audio",
                  data: base64Data,
                  mimeType,
              },
              createTextContent(
                  `Generated audio for text: "${text}"\n\nVoice: ${voice}\nFormat: ${format}`,
              ),
          ]);
      } catch (error) {
          console.error("Error generating audio:", error);
          throw error;
      }
  }

• src/services/audioService.js:359-378 (schema)

  Input schema using Zod for validating parameters of the sayText tool: text (required), voice, format, voiceInstructions (optional).

  {
      text: z.string().describe("The text to speak verbatim"),
      voice: z
          .string()
          .optional()
          .describe(
              'Voice to use for audio generation (default: "alloy")',
          ),
      format: z
          .string()
          .optional()
          .describe("Format of the audio (mp3, wav, etc.)"),
      voiceInstructions: z
          .string()
          .optional()
          .describe(
              'Additional instructions for voice character/style (e.g., "Speak with enthusiasm" or "Use a calm tone")',
          ),
  },
  sayText,

• src/services/audioService.js:356-379 (registration)

  Registration entry for the sayText tool in the audioTools export array, formatted for MCP server.tool() calls.

  [
      "sayText",
      "Generate speech that says the provided text verbatim",
      {
          text: z.string().describe("The text to speak verbatim"),
          voice: z
              .string()
              .optional()
              .describe(
                  'Voice to use for audio generation (default: "alloy")',
              ),
          format: z
              .string()
              .optional()
              .describe("Format of the audio (mp3, wav, etc.)"),
          voiceInstructions: z
              .string()
              .optional()
              .describe(
                  'Additional instructions for voice character/style (e.g., "Speak with enthusiasm" or "Use a calm tone")',
              ),
      },
      sayText,
  ],