ask-ollama-mcp

Ask Ollama MCP

MCP server for local Ollama LLMs — no API keys, fully private

An MCP server for AI-to-AI collaboration via local Ollama models. Works with Claude Code, Claude Desktop, Cursor, Warp, Copilot, and 40+ other MCP clients. Run code reviews and analysis entirely locally — no data leaves your machine, no API keys needed, zero cost.

Part of the Ask LLM monorepo.

Quick Start

Claude Code

claude mcp add ollama -- npx -y ask-ollama-mcp

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "ollama": {
      "command": "npx",
      "args": ["-y", "ask-ollama-mcp"]
    }
  }
}

Any MCP Client

{
  "command": "npx",
  "args": ["-y", "ask-ollama-mcp"]
}

Prerequisites

• Node.js v20.0.0 or higher
• Ollama installed and running locally
• A model pulled: ollama pull qwen3.6:27b

Tools

Tool	Purpose
ask-ollama	Send prompts to local Ollama via HTTP. Defaults to qwen3.6:27b
ping	Lists locally available Ollama models via /api/tags

Models

Model	Use Case
qwen3.6:27b	Default — Qwen's flagship-level local coding model (~17 GB)

Override with ASK_OLLAMA_MODEL. Ollama is local, so there's no automatic fallback — you pull the model you want, and a missing model returns a clear ollama pull <model> error.

Configuration

Set OLLAMA_HOST to customize the Ollama server address (default: http://localhost:11434).

Documentation

Full docs at lykhoyda.github.io/ask-llm

License

MIT