Ollama MCP Server

An Ollama model collection can be inspected and maintained from a connected assistant: ollama_list reports installed models, while ollama_ps identifies those loaded in memory. Users can examine an individual model's configuration and architecture, download models from the registry, upload local models, copy them under new names, and delete them locally. Custom model creation supports changes to behavior, system prompts, and templates. For inference, ollama_chat handles conversations and tool calling, ollama_generate handles prompt completions, and ollama_embed turns text into vectors.

Ollama Cloud adds web search with a configurable result limit and page retrieval that supplies titles, content, and links. These operations and cloud models require OLLAMA_API_KEY. For hybrid use, the configuration targets a local Ollama instance and includes the key, allowing local inference alongside cloud web tools. Running on Node.js with stdio transport, ollama-mcp connects these workflows to compatible clients, including Claude Desktop and Cline.