Flama

What it does

Flama is a high-performance production framework designed to turn any predictive or generative AI model into a production-ready API in a single line of code. It allows developers to serve models from various frameworks (scikit-learn, TensorFlow, PyTorch, LLMs) using a unified portable .flm format, exposing them via OpenAI, Anthropic, and Ollama-compatible endpoints.

Tools

Flama enables the creation of custom MCP tools via a simple Python decorator. By adding a server to the Flama app, any function marked with @app.mcp.tool is automatically exposed as an MCP tool with a JSON Schema derived from its type hints.

Installation

To install Flama and its generative AI capabilities:

pip install "flama[llm,pydantic]"

To serve a model as an MCP server:

flama serve --model file=your_model.flm,url=/,name=model_name

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "flama": {
      "command": "flama",
      "args": ["serve", "--model", "file=your_model.flm,url=/,name=model_name"]
    }
  }
}

Supported hosts

• claude
• cursor
• vscode-copilot