An MCP (Model Context Protocol) server that enables AI applications to outsource tasks to various model providers through a unified interface.

Compatible with any AI tool that supports the Model Context Protocol, including Claude Desktop, Cline, and other MCP-enabled applications. Built with FastMCP for the MCP server implementation and Agno for AI agent capabilities.

• 🤖 Multi-Provider Support: Access 20+ AI providers through a single interface
• 📝 Text Generation: Generate text using models from OpenAI, Anthropic, Google, and more
• 🎨 Image Generation: Create images using DALL-E 3 and DALL-E 2
• 🔧 Simple API: Consistent interface with just three parameters: provider, model, and prompt
• 🔑 Flexible Authentication: Only configure API keys for the providers you use

Add the following configuration to your MCP client. Consult your MCP client's documentation for specific configuration details.

{
  "mcpServers": {
    "outsource-mcp": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/gwbischof/outsource-mcp.git", "outsource-mcp"],
      "env": {
        "OPENAI_API_KEY": "your-openai-key",
        "ANTHROPIC_API_KEY": "your-anthropic-key",
        "GOOGLE_API_KEY": "your-google-key",
        "GROQ_API_KEY": "your-groq-key",
        "DEEPSEEK_API_KEY": "your-deepseek-key",
        "XAI_API_KEY": "your-xai-key",
        "PERPLEXITY_API_KEY": "your-perplexity-key",
        "COHERE_API_KEY": "your-cohere-key",
        "FIREWORKS_API_KEY": "your-fireworks-key",
        "HUGGINGFACE_API_KEY": "your-huggingface-key",
        "MISTRAL_API_KEY": "your-mistral-key",
        "NVIDIA_API_KEY": "your-nvidia-key",
        "OLLAMA_HOST": "https://localhost:11434",
        "OPENROUTER_API_KEY": "your-openrouter-key",
        "TOGETHER_API_KEY": "your-together-key",
        "CEREBRAS_API_KEY": "your-cerebras-key",
        "DEEPINFRA_API_KEY": "your-deepinfra-key",
        "SAMBANOVA_API_KEY": "your-sambanova-key"
      }
    }
  }
}

Note: The environment variables are optional. Only include the API keys for the providers you want to use.

Once installed and configured, you can use the tools in your MCP client:

1. Generate text: Use the outsource_text tool with provider "openai", model "gpt-4o-mini", and prompt "Write a haiku about coding"
2. Generate images: Use the outsource_image tool with provider "openai", model "dall-e-3", and prompt "A futuristic city skyline at sunset"

Creates an Agno agent with a specified provider and model to generate text responses.

Arguments:

• provider: The provider name (e.g., "openai", "anthropic", "google", "groq", etc.)
• model: The model name (e.g., "gpt-4o", "claude-3-5-sonnet-20241022", "gemini-2.0-flash-exp")
• prompt: The text prompt to send to the model

Generates images using AI models.

Arguments:

• provider: The provider name (currently only "openai" is supported)
• model: The model name ("dall-e-3" or "dall-e-2")
• prompt: The image generation prompt

Returns the URL of the generated image.

Note: Image generation is currently only supported by OpenAI models (DALL-E 2 and DALL-E 3). Other providers only support text generation.

The following providers are supported. Use the provider name (in parentheses) as the provider argument:

• OpenAI (openai) - GPT-4, GPT-3.5, DALL-E, etc. | Models
• Anthropic (anthropic) - Claude 3.5, Claude 3, etc. | Models
• Google (google) - Gemini Pro, Gemini Flash, etc. | Models
• Groq (groq) - Llama 3, Mixtral, etc. | Models
• DeepSeek (deepseek) - DeepSeek Chat & Coder | Models
• xAI (xai) - Grok models | Models
• Perplexity (perplexity) - Sonar models | Models

• Cohere (cohere) - Command models | Models
• Mistral AI (mistral) - Mistral Large, Medium, Small | Models
• NVIDIA (nvidia) - Various models | Models
• HuggingFace (huggingface) - Open source models | Models
• Ollama (ollama) - Local models | Models
• Fireworks AI (fireworks) - Fast inference | Models
• OpenRouter (openrouter) - Multi-provider access | Models
• Together AI (together) - Open source models | Models
• Cerebras (cerebras) - Fast inference | Models
• DeepInfra (deepinfra) - Optimized models | Models
• SambaNova (sambanova) - Enterprise models | Models

• AWS Bedrock (aws or bedrock) - AWS-hosted models | Models
• Azure AI (azure) - Azure-hosted models | Models
• IBM WatsonX (ibm or watsonx) - IBM models | Models
• LiteLLM (litellm) - Universal interface | Models
• Vercel v0 (vercel or v0) - Vercel AI | Models
• Meta Llama (meta) - Direct Meta access | Models

Each provider requires its corresponding API key:

Note: Only configure the API keys for providers you plan to use.

# Using OpenAI
provider: openai
model: gpt-4o-mini
prompt: Write a haiku about coding
# Using Anthropic
provider: anthropic
model: claude-3-5-sonnet-20241022
prompt: Explain quantum computing in simple terms
# Using Google
provider: google
model: gemini-2.0-flash-exp
prompt: Create a recipe for chocolate chip cookies

# Using DALL-E 3
provider: openai
model: dall-e-3
prompt: A serene Japanese garden with cherry blossoms
# Using DALL-E 2
provider: openai
model: dall-e-2
prompt: A futuristic cityscape at sunset

• Python 3.11 or higher
• uv package manager

git clone https://github.com/gwbischof/outsource-mcp.git
cd outsource-mcp
uv sync

The MCP Inspector allows you to test the server interactively:

mcp dev server.py

The test suite includes integration tests that verify both text and image generation:

# Run all tests
uv run pytest

Note: Integration tests require API keys to be set in your environment.

1. "Error: Unknown provider"

   • Check that you're using a supported provider name from the list above
   • Provider names are case-insensitive

2. "Error: OpenAI API error"

   • Verify your API key is correctly set in the environment variables
   • Check that your API key has access to the requested model
   • Ensure you have sufficient credits/quota

3. "Error: No image was generated"

   • This can happen if the image generation request fails
   • Try a simpler prompt or different model (dall-e-2 vs dall-e-3)

4. Environment variables not working

   • Make sure to restart your MCP client after updating the configuration
   • Verify the configuration file location for your specific MCP client
   • Check that the environment variables are properly formatted in the configuration

Contributions are welcome! Please feel free to submit a Pull Request.