MCP DeepInfra AI Tools Server

This is a Model Context Protocol (MCP) server that provides various AI capabilities using the DeepInfra OpenAI-compatible API, including image generation, text processing, embeddings, speech recognition, and more.

Project Structure

mcp-deepinfra/ ├── src/ │ └── mcp_deepinfra/ │ ├── __init__.py # Package initialization │ └── server.py # Main MCP server implementation ├── tests/ │ ├── conftest.py # Pytest fixtures and configuration │ ├── test_server.py # Server initialization tests │ └── test_tools.py # Individual tool tests ├── pyproject.toml # Project configuration and dependencies ├── uv.lock # Lock file for uv package manager ├── run_tests.sh # Convenience script for running tests └── README.md # This file

Setup

Install uv if not already installed:
curl -LsSf https://astral.sh/uv/install.sh | sh
Clone or download this repository.
Install dependencies:
uv sync
Set up your DeepInfra API key: Create a .env file in the project root:
DEEPINFRA_API_KEY=your_api_key_here

Configuration

You can configure which tools are enabled and set default models for each tool using environment variables in your .env file:

ENABLED_TOOLS: Comma-separated list of tools to enable. Use "all" to enable all tools (default: "all"). Example: ENABLED_TOOLS=generate_image,text_generation,embeddings
MODEL_GENERATE_IMAGE: Default model for image generation (default: "Bria/Bria-3.2")
MODEL_TEXT_GENERATION: Default model for text generation (default: "meta-llama/Llama-2-7b-chat-hf")
MODEL_EMBEDDINGS: Default model for embeddings (default: "sentence-transformers/all-MiniLM-L6-v2")
MODEL_SPEECH_RECOGNITION: Default model for speech recognition (default: "openai/whisper-large-v3")
MODEL_ZERO_SHOT_IMAGE_CLASSIFICATION: Default model for zero-shot image classification (default: "openai/gpt-4o-mini")
MODEL_OBJECT_DETECTION: Default model for object detection (default: "openai/gpt-4o-mini")
MODEL_IMAGE_CLASSIFICATION: Default model for image classification (default: "openai/gpt-4o-mini")
MODEL_TEXT_CLASSIFICATION: Default model for text classification (default: "microsoft/DialoGPT-medium")
MODEL_TOKEN_CLASSIFICATION: Default model for token classification (default: "microsoft/DialoGPT-medium")
MODEL_FILL_MASK: Default model for fill mask (default: "microsoft/DialoGPT-medium")

The tools always use the models specified via environment variables. Model selection is configured at startup time through the environment variables listed above.

Running the Server

To run the server locally:

uv run mcp_deepinfra

Or directly with Python:

python -m mcp_deepinfra.server

Using with MCP Clients

Configure your MCP client (e.g., Claude Desktop) to use this server.

For Claude Desktop, add to your claude_desktop_config.json:

{ "mcpServers": { "deepinfra": { "command": "uv", "args": ["run", "mcp_deepinfra"], "env": { "DEEPINFRA_API_KEY": "your_api_key_here" } } } }

Tools Provided

This server provides the following MCP tools:

generate_image: Generate an image from a text prompt. Returns the URL of the generated image.
text_generation: Generate text completion from a prompt.
embeddings: Generate embeddings for a list of input texts.
speech_recognition: Transcribe audio from a URL to text using Whisper model.
zero_shot_image_classification: Classify an image into provided candidate labels using vision model.
object_detection: Detect and describe objects in an image using multimodal model.
image_classification: Classify and describe contents of an image using multimodal model.
text_classification: Analyze text for sentiment and category.
token_classification: Perform named entity recognition (NER) on text.
fill_mask: Fill masked tokens in text with appropriate words.

Testing

To test the server locally, run the pytest test suite:

# Install test dependencies uv sync --extra test # Run all tests pytest # Run with verbose output pytest -v # Run specific test file pytest tests/test_tools.py # Use the convenience script ./run_tests.sh

The tests include: