MCP DeepInfra AI Tools Server

Instructions

Implementation Reference

• src/mcp_deepinfra/server.py:121-150 (handler)

  The handler function that executes the zero_shot_image_classification tool. It constructs a vision-enabled chat completion request to classify the image from the given candidate labels using the configured model.

  async def zero_shot_image_classification(image_url: str, candidate_labels: list[str]) -> str:
      """Classify an image with zero-shot labels using DeepInfra OpenAI-compatible API (CLIP)."""
      model = DEFAULT_MODELS["zero_shot_image_classification"]
      try:
          # Use chat/completions with vision capability to get classification
          response = await client.chat.completions.create(
              model=model,
              messages=[
                  {
                      "role": "user",
                      "content": [
                          {
                              "type": "text",
                              "text": f"Classify this image into one of these categories: {', '.join(candidate_labels)}. Return a JSON with 'label' and 'score' fields."
                          },
                          {
                              "type": "image_url",
                              "image_url": {"url": image_url}
                          }
                      ]
                  }
              ],
              max_tokens=200,
          )
          if response.choices:
              return response.choices[0].message.content
          else:
              return "Unable to classify image"
      except Exception as e:
          return f"Error classifying image: {type(e).__name__}: {str(e)}"

• src/mcp_deepinfra/server.py:119-120 (registration)

  Conditional registration of the zero_shot_image_classification tool using the FastMCP @app.tool() decorator.

  if "all" in ENABLED_TOOLS or "zero_shot_image_classification" in ENABLED_TOOLS:
      @app.tool()

• src/mcp_deepinfra/server.py:121-122 (schema)

  Function signature providing input schema (image_url: str, candidate_labels: list[str]) and output str, along with docstring description used for tool schema in MCP.

  async def zero_shot_image_classification(image_url: str, candidate_labels: list[str]) -> str:
      """Classify an image with zero-shot labels using DeepInfra OpenAI-compatible API (CLIP)."""

• src/mcp_deepinfra/server.py:36-36 (helper)

  Helper configuration defining the default model for the zero_shot_image_classification tool.

  "zero_shot_image_classification": os.getenv("MODEL_ZERO_SHOT_IMAGE_CLASSIFICATION", "openai/gpt-4o-mini"),