add-ollama-tool

You can run local model inference inside a container agent by enabling an MCP server that bridges local Ollama models to the agent, allowing it to offload tasks for faster and cheaper processing.

Yes, you can use Ollama models with an MCP server to achieve cheaper and faster on-host inference, enabling local tasks such as general queries, translation, or summarization without sending data to external services.

Before querying local Ollama models, you need Python3 installed, Ollama running on the host with at least one model installed, and you must apply skill changes via Phase 2 and restart MatClaw.

You list and generate responses from local models by using the provided ollama_list_models tool to inspect available models and the ollama_generate tool to locally query a model for your tasks.

The best way to accelerate code tasks without external data transmission is to configure an MCP server bridging local Ollama models to your container agent, executing on-host model inference locally.

The Ollama MCP server fails to list models if Ollama is not installed or running on the host, if no model is installed, or if the OLLAMA_HOST configuration is incorrect during Phase 2 setup before restarting MatClaw.