add-ollama-tool

To expose local Ollama models to container agents, you can use a stdio MCP server integration. This enables on-host, private inference for tasks like summarization and translation without sending data to external servers.

Yes, you can run local model summarization tasks inside isolated containers using MCP. This setup provides host isolation and simple host discovery while keeping data private during on-host inference.

The available MCP tools for integrating Ollama with agent swarms are ollama_list_models and ollama_generate. These tools allow agents to discover available local models and execute inference tasks directly.

Yes, you need to configure OLLAMA_HOST when using Ollama MCP tools in container agents. The stdio MCP server uses this configurable host setting to enable simple host discovery and secure local model inference.

Yes, this approach supports code reasoning tasks without sending data to external servers. By exposing Ollama models via an MCP server, container agents can perform fast, private in-host model tasks securely.

Using local Ollama models for agent inference in containers requires on-host availability and proper OLLAMA_HOST configuration. Performance depends on local hardware resources, and inference is limited to the models installed locally.