Offline MCP

What it does

Offline MCP provides a sovereignty floor for local AI inference, specifically designed for environments with limited connectivity. It wraps Ollama and open-weight models to ensure health, civic, and land data remains local and secure from foreign extraction.

Tools

• check_ollama_status: Verifies if Ollama is running and lists available models.
• run_local_inference: Executes prompts through a local Ollama model.
• list_recommended_models: Suggests best open-weight models for East African use cases.
• degraded_mode_guide: Provides a 4-level architecture for offline operational stability.
• open_weights_directory: A directory of models with African language support.
• local_deployment_guide: Guides deployment on Raspberry Pi, Android, and servers.

Installation

Install via pip:

pip install offline-mcp

Then configure your MCP client to point to the offline-mcp server implementation.

Supported hosts

General MCP compatibility; designed for local deployments (e.g., Raspberry Pi).