Summary
Dropping to a terminal to pull a model, then again to check what is still loaded.
Inference and library management are the same tool list, so trying a model you do not have yet is pull then generate rather than a context switch. Two of the thirteen tools leave the machine: ollama_web_search and ollama_web_fetch call Ollama's hosted APIs and want OLLAMA_API_KEY, which is worth noticing on a server you installed to keep things local.
What it is
An Ollama client with thirteen tools: inference against local models, management of the local model library, and two tools that reach Ollama's hosted web search and fetch APIs.
What you get
- Multi-turn chat with system messages, tool calling and generation options, and a simpler single-turn completion alongside it
- Embeddings for text input, returned as numerical vectors
- The local library managed in place: list what is installed with sizes and modification dates, pull, delete, copy and push
- New models created with their own system prompt, template, license and base model
- Detailed model information — modelfile, parameters and architecture
- The models currently loaded in memory, separately from the ones on disk
- Web search and page fetch through Ollama's hosted APIs, returning results, or a page's title, content and links
Requirements
An Ollama installation with the models you intend to call. ollama_web_search and ollama_web_fetch read OLLAMA_API_KEY from the environment.
Setup effort
One command — npx -y ollama-mcp