add-ollama-tool

This skill provides an MCP server that exposes local Ollama models as agent tools for container-based environments. It enables on-device model inference, allowing Docker agents to generate text without relying on remote endpoints.

Yes, you can manage Ollama models and memory from a container agent by enabling OLLAMA_ADMIN_TOOLS. This opt-in setting unlocks management tools to pull, delete, and inspect models, as well as list running models and memory usage.

Yes, you need Ollama installed and running on the host machine to use local models with a container agent. The container agent communicates with the host's Ollama instance to perform on-device inference and optional model management.

Available MCP agent features include listing installed models, generating responses, and optionally pulling or deleting models. Enabling admin tools also allows agents to view model details, list running models, and check memory usage.

Using local Ollama models instead of remote endpoints for AI agent inference ensures on-device processing within container environments. This approach reduces reliance on external APIs, keeping model execution and data handling entirely local.