What problem does it solve? Containerized agents normally cannot reach local LLMs running on the host machine. This Skill wires a stdio-based MCP server into the NanoClaw agent container so Claude can offload work to local Ollama models and optionally manage the model library, without threading any credentials. ## Core Features & Use Cases - Local model inference tools: Exposes ollama_list_models and ollama_generate so the agent can list installed models and send prompts to them via the Ollama REST API. - Optional library management: With OLLAMA_ADMIN_TOOLS=true, adds tools to pull, delete, inspect, and list running models on the host daemon. - Guided installation with verification: Copies the MCP server source into the container and host trees, registers it in index.ts, forwards env vars, and validates the wiring with structural tests. - Use Case: You want your agent to summarize long documents cheaply using a local llama3.2 model instead of paid API calls, while keeping the option to pull new models on demand. ## Quick Start Ask the agent to add the Ollama tool so it can list local models and generate responses with them.
