MCP DeepInfra AI Tools Server

Instructions

Implementation Reference

• src/mcp_deepinfra/server.py:98-118 (handler)

  The speech_recognition tool handler, conditionally defined and registered via @app.tool() decorator. It downloads audio from a provided URL, then uses DeepInfra's OpenAI-compatible Whisper API for transcription.

  if "all" in ENABLED_TOOLS or "speech_recognition" in ENABLED_TOOLS:
      @app.tool()
      async def speech_recognition(audio_url: str) -> str:
          """Transcribe audio to text using DeepInfra OpenAI-compatible API (Whisper)."""
          model = DEFAULT_MODELS["speech_recognition"]
          try:
              async with httpx.AsyncClient(timeout=120.0) as http_client:
                  # Download the audio file
                  audio_response = await http_client.get(audio_url)
                  audio_response.raise_for_status()
                  audio_content = audio_response.content
              
              # Use the OpenAI-compatible Whisper API
              response = await client.audio.transcriptions.create(
                  model=model,
                  file=("audio.mp3", audio_content),
              )
              return response.text
          except Exception as e:
              return f"Error transcribing audio: {type(e).__name__}: {str(e)}"

• src/mcp_deepinfra/server.py:31-42 (helper)

  Configuration dictionary DEFAULT_MODELS defining the default model for speech_recognition tool (Whisper large-v3). Used within the handler.

  DEFAULT_MODELS = {
      "generate_image": os.getenv("MODEL_GENERATE_IMAGE", "Bria/Bria-3.2"),
      "text_generation": os.getenv("MODEL_TEXT_GENERATION", "meta-llama/Llama-2-7b-chat-hf"),
      "embeddings": os.getenv("MODEL_EMBEDDINGS", "sentence-transformers/all-MiniLM-L6-v2"),
      "speech_recognition": os.getenv("MODEL_SPEECH_RECOGNITION", "openai/whisper-large-v3"),
      "zero_shot_image_classification": os.getenv("MODEL_ZERO_SHOT_IMAGE_CLASSIFICATION", "openai/gpt-4o-mini"),
      "object_detection": os.getenv("MODEL_OBJECT_DETECTION", "openai/gpt-4o-mini"),
      "image_classification": os.getenv("MODEL_IMAGE_CLASSIFICATION", "openai/gpt-4o-mini"),
      "text_classification": os.getenv("MODEL_TEXT_CLASSIFICATION", "microsoft/DialoGPT-medium"),
      "token_classification": os.getenv("MODEL_TOKEN_CLASSIFICATION", "microsoft/DialoGPT-medium"),
      "fill_mask": os.getenv("MODEL_FILL_MASK", "microsoft/DialoGPT-medium"),
  }