add-ollama-tool

You expose local Ollama models to a container agent by adding a stdio-based MCP server. This server provides tools like ollama_list_models and ollama_generate, allowing the agent to offload offline inference tasks such as summarization and coding locally.

A stdio-based MCP server for local AI integration is an interface that connects container agents to local models using standard input and output. It allows Claude to act as an orchestrator while offloading work to local Ollama models.

Yes, you need Ollama installed to use local models for offline container tasks. The guided workflow expects an existing Ollama setup to enable the MCP tool and allow the container agent to call local models for offline inference.

The best way to run offline model inference in an isolated sandbox is using a guided pre-flight, apply, configure, and verify workflow. This exposes Ollama models via an MCP server, allowing container agents to call local models securely.

Yes, you can configure a custom Ollama host for your containerized MCP tools. The server supports optional Ollama host configuration during the apply and configure phases, ensuring the agent connects to the correct local model endpoint.

You use local models instead of the container agent for coding tasks to offload processing work. Claude remains the orchestrator but delegates specific offline inference tasks to Ollama models, optimizing resource usage within isolated environments.