mcp_openai_transcribe

Instructions

Implementation Reference

• src/tools/index.ts:742-759 (handler)

  The inline handler function for the mcp_openai_transcribe tool. It delegates to openaiService.speechToText and handles the response or error.

  async handler(args: any): Promise<ToolResponse> {
    try {
      const result = await openaiService.speechToText(args);
      return {
        content: [{
          type: 'text',
          text: result
        }]
      };
    } catch (error) {
      return {
        content: [{
          type: 'text',
          text: `OpenAI Whisper 오류: ${error instanceof Error ? error.message : String(error)}`
        }]
      };
    }
  }

• src/tools/index.ts:720-741 (schema)

  The input schema defining parameters for the transcription tool: audioPath (required), model, language, prompt.

  inputSchema: {
    type: 'object',
    properties: {
      audioPath: {
        type: 'string',
        description: '변환할 오디오 파일 경로'
      },
      model: {
        type: 'string',
        description: '사용할 모델 (예: whisper-1)'
      },
      language: {
        type: 'string',
        description: '오디오 언어 (예: ko, en, ja)'
      },
      prompt: {
        type: 'string',
        description: '인식을 도울 힌트 텍스트'
      }
    },
    required: ['audioPath']
  },

• src/index.ts:24-54 (registration)

  MCP server capabilities registration declaring mcp_openai_transcribe as available (true). The server uses imported tools array to handle calls.

  capabilities: {
    tools: {
      mcp_sparql_execute_query: true,
      mcp_sparql_update: true,
      mcp_sparql_list_repositories: true,
      mcp_sparql_list_graphs: true,
      mcp_sparql_get_resource_info: true,
      mcp_ollama_run: true,
      mcp_ollama_show: true,
      mcp_ollama_pull: true,
      mcp_ollama_list: true,
      mcp_ollama_rm: true,
      mcp_ollama_chat_completion: true,
      mcp_ollama_status: true,
      mcp_http_request: true,
      mcp_openai_chat: true,
      mcp_openai_image: true,
      mcp_openai_tts: true,
      mcp_openai_transcribe: true,
      mcp_openai_embedding: true,
      mcp_gemini_generate_text: true,
      mcp_gemini_chat_completion: true,
      mcp_gemini_list_models: true,
      mcp_gemini_generate_images: false,
      mcp_gemini_generate_image: false,
      mcp_gemini_generate_videos: false,
      mcp_gemini_generate_multimodal_content: false,
      mcp_imagen_generate: false,
      mcp_gemini_create_image: false,
      mcp_gemini_edit_image: false
    },

• src/services/openai-service.ts:255-318 (helper)

  The core speechToText method in OpenAIService that performs the actual OpenAI Whisper API call for audio transcription using multipart form data.

  async speechToText(args: {
    audioPath: string;
    model?: string;
    language?: string;
    prompt?: string;
  }): Promise<string> {
    try {
      if (!OPENAI_API_KEY) {
        throw new McpError(
          ErrorCode.InternalError, 
          'OPENAI_API_KEY가 설정되지 않았습니다.'
        );
      }
  
      if (!fs.existsSync(args.audioPath)) {
        throw new McpError(
          ErrorCode.InternalError,
          `오디오 파일을 찾을 수 없습니다: ${args.audioPath}`
        );
      }
  
      const formData = new FormData();
      const fileBlob = new Blob([fs.readFileSync(args.audioPath)]);
      formData.append('file', fileBlob, path.basename(args.audioPath));
      formData.append('model', args.model || 'whisper-1');
      
      if (args.language) {
        formData.append('language', args.language);
      }
      
      if (args.prompt) {
        formData.append('prompt', args.prompt);
      }
  
      const response = await axios.post(
        `${OPENAI_API_BASE}/audio/transcriptions`,
        formData,
        {
          headers: {
            'Content-Type': 'multipart/form-data',
            'Authorization': `Bearer ${OPENAI_API_KEY}`
          }
        }
      );
  
      return JSON.stringify(response.data, null, 2);
    } catch (error) {
      if (axios.isAxiosError(error)) {
        const statusCode = error.response?.status;
        const responseData = error.response?.data;
        
        throw new McpError(
          ErrorCode.InternalError,
          `OpenAI API 오류 (${statusCode}): ${
            typeof responseData === 'object' 
              ? JSON.stringify(responseData, null, 2) 
              : responseData || error.message
          }`
        );
      }
      
      throw new McpError(ErrorCode.InternalError, `음성 인식 요청 실패: ${formatError(error)}`);
    }
  }