Blabber-MCP

#!/usr/bin/env node import { Server } from "@modelcontextprotocol/sdk/server/index.js"; import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js"; import { CallToolRequestSchema, ListToolsRequestSchema, McpError, ErrorCode, } from "@modelcontextprotocol/sdk/types.js"; import OpenAI from "openai"; import { APIError } from "openai/error.js"; import fs from 'fs'; import path from 'path'; import { fileURLToPath } from 'url'; import { exec } from 'child_process'; // --- Configuration --- const API_KEY = process.env.OPENAI_API_KEY; if (!API_KEY) { console.error("Error: OPENAI_API_KEY environment variable is not set."); process.exit(1); } const AUDIO_PLAYER_COMMAND = process.env.AUDIO_PLAYER_COMMAND || 'xdg-open'; // Define allowed voices const ALLOWED_VOICES = ["alloy", "echo", "fable", "onyx", "nova", "shimmer"] as const; type AllowedVoice = typeof ALLOWED_VOICES[number]; // Read default voice from env var, validate, and set default let DEFAULT_VOICE: AllowedVoice = "nova"; const configuredDefaultVoice = process.env.DEFAULT_TTS_VOICE; if (configuredDefaultVoice && (ALLOWED_VOICES as readonly string[]).includes(configuredDefaultVoice)) { DEFAULT_VOICE = configuredDefaultVoice as AllowedVoice; console.error(`Using configured default voice: ${DEFAULT_VOICE}`); } else if (configuredDefaultVoice) { console.error(`Warning: Invalid DEFAULT_TTS_VOICE "${configuredDefaultVoice}" provided. Using default "alloy".`); } else { console.error(`Using default voice: ${DEFAULT_VOICE}`); } const __filename = fileURLToPath(import.meta.url); const __dirname = path.dirname(__filename); const OUTPUT_DIR = path.resolve(__dirname, '..', 'output'); const openai = new OpenAI({ apiKey: API_KEY, }); // --- MCP Server Setup --- const server = new Server( { name: "@pink/pixel/blabber-mcp", version: "0.1.2", }, { capabilities: { tools: {}, }, } ); // --- Tool Definition --- const TEXT_TO_SPEECH_TOOL_NAME = "text_to_speech"; server.setRequestHandler(ListToolsRequestSchema, async () => { return { tools: [ { name: TEXT_TO_SPEECH_TOOL_NAME, description: `Converts text into spoken audio using OpenAI TTS (default voice: ${DEFAULT_VOICE}), saves it to a file, and optionally plays it.`, // Updated description inputSchema: { type: "object", properties: { input: { type: "string", description: "The text to synthesize into speech.", }, voice: { type: "string", description: `Optional: The voice to use. Overrides the configured default (${DEFAULT_VOICE}).`, enum: [...ALLOWED_VOICES], // Use the defined constant }, model: { type: "string", description: "The TTS model to use.", enum: ["tts-1", "tts-1-hd"], default: "tts-1", }, response_format: { type: "string", description: "The format of the audio response.", enum: ["mp3", "opus", "aac", "flac"], default: "mp3", }, play: { type: "boolean", description: "Whether to automatically play the generated audio file.", default: false, } }, required: ["input"], }, }, ], }; }); // --- Tool Implementation --- type TextToSpeechArgs = { input: string; voice?: AllowedVoice; // Use the specific type model?: "tts-1" | "tts-1-hd"; response_format?: "mp3" | "opus" | "aac" | "flac"; play?: boolean; }; // Updated type guard function isValidTextToSpeechArgs(args: any): args is TextToSpeechArgs { return ( typeof args === "object" && args !== null && typeof args.input === "string" && (args.voice === undefined || (ALLOWED_VOICES as readonly string[]).includes(args.voice)) && // Validate against allowed voices (args.model === undefined || ["tts-1", "tts-1-hd"].includes(args.model)) && (args.response_format === undefined || ["mp3", "opus", "aac", "flac"].includes(args.response_format)) && (args.play === undefined || typeof args.play === 'boolean') ); } server.setRequestHandler(CallToolRequestSchema, async (request) => { if (request.params.name !== TEXT_TO_SPEECH_TOOL_NAME) { throw new McpError(ErrorCode.MethodNotFound, `Unknown tool: ${request.params.name}`); } if (!isValidTextToSpeechArgs(request.params.arguments)) { throw new McpError(ErrorCode.InvalidParams, "Invalid arguments for text_to_speech tool."); } const { input, // Use voice from args if provided, otherwise use the configured DEFAULT_VOICE voice = DEFAULT_VOICE, model = "tts-1", response_format = "mp3", play = false, } = request.params.arguments; // Ensure the final voice is valid (handles case where default might somehow be invalid, though unlikely with validation above) const finalVoice: AllowedVoice = (ALLOWED_VOICES as readonly string[]).includes(voice) ? voice : DEFAULT_VOICE; let playbackMessage = ""; try { if (!fs.existsSync(OUTPUT_DIR)) { fs.mkdirSync(OUTPUT_DIR, { recursive: true }); console.error(`Created output directory: ${OUTPUT_DIR}`); } console.error(`Generating speech with voice: ${finalVoice}`); // Log the voice being used const speechResponse = await openai.audio.speech.create({ model: model, voice: finalVoice, // Use the validated final voice input: input, response_format: response_format, }); const audioBuffer = Buffer.from(await speechResponse.arrayBuffer()); const timestamp = Date.now(); const filename = `speech_${timestamp}.${response_format}`; const filePath = path.join(OUTPUT_DIR, filename); const relativeFilePath = path.relative(process.cwd(), filePath); fs.writeFileSync(filePath, audioBuffer); console.error(`Audio saved to: ${filePath}`); if (play) { const command = `${AUDIO_PLAYER_COMMAND} "${filePath}"`; console.error(`Attempting to play audio with command: ${command}`); exec(command, (error, stdout, stderr) => { if (error) console.error(`Playback Error: ${error.message}`); if (stderr) console.error(`Playback Stderr: ${stderr}`); if (stdout) console.error(`Playback stdout: ${stdout}`); }); playbackMessage = ` Playback initiated using command: ${AUDIO_PLAYER_COMMAND}.`; } return { content: [ { type: "text", text: JSON.stringify({ message: `Audio saved successfully.${playbackMessage}`, filePath: relativeFilePath, format: response_format, voiceUsed: finalVoice, // Inform client which voice was actually used }), mimeType: "application/json", }, ], }; } catch (error) { let errorMessage = "Failed to generate speech."; if (error instanceof APIError) { errorMessage = `OpenAI API Error (${error.status}): ${error.message}`; } else if (error instanceof Error) { errorMessage = error.message; } console.error(`[${TEXT_TO_SPEECH_TOOL_NAME} Error]`, errorMessage, error); return { content: [{ type: "text", text: errorMessage }], isError: true } } }); // --- Server Start --- async function main() { const transport = new StdioServerTransport(); server.onerror = (error) => console.error("[MCP Error]", error); process.on('SIGINT', async () => { console.error("Received SIGINT, shutting down server..."); await server.close(); process.exit(0); }); await server.connect(transport); console.error("OpenAI TTS MCP server running on stdio"); } main().catch((error) => { console.error("Server failed to start:", error); process.exit(1); });