Ollama MCP (Enhanced)

Overview

Ollama MCP provides access to locally-running Ollama models for private AI inference.

Features

• Local model inference
• Multiple model support
• Streaming responses
• Model management
• GPU acceleration

Installation

npx -y @modelcontextprotocol/server-ollama

Configuration

Set up Ollama host:

export OLLAMA_HOST=http://localhost:11434

Pull models with: ollama pull llama3

Available Models

Model	Description
llama3	Meta Llama 3
mistral	Mistral 7B
gemma	Google Gemma
codellama	Code generation
phi3	Microsoft Phi 3

Available Operations

Operation	Description
Chat	Send chat messages
Generate	Text generation
Embed	Generate embeddings
List Models	Get available models
Pull Model	Download model

Usage Examples

Chat

{
  "action": "chat",
  "model": "llama3",
  "messages": [
    { "role": "user", "content": "Explain quantum computing" }
  ],
  "stream": true
}

Generate

{
  "action": "generate",
  "model": "llama3",
  "prompt": "Write a Python function to calculate factorial"
}

Embed

{
  "action": "embed",
  "model": "llama3",
  "input": ["Hello world", "AI is amazing"]
}

Pros

• ✅ Fully local and private
• ✅ No API costs
• ✅ GPU acceleration
• ✅ Model variety

Cons

• ❌ Requires local Ollama
• ❌ Hardware requirements
• ❌ Model download time

Resources

• Ollama Docs
• MCP Server GitHub