Ollama Turbo Desktop

A refactored version of Groq Desktop for use with Ollama Turbo cloud inference (or local). The MCP implementaton is very well done and excels at MCP tool use. Built with Electron, React, and the latest MCP SDK for seamless tool integration.

Available for Windows, macOS, and Linux!

Note Groq Desktop provides an installable version while this is a dev experiment demonstrated on Episode 12 of The Build Podcast:

Features

• Accelerated AI Models: Chat with fast GPT-OSS models (20B and 120B parameters) via Ollama Turbo
• MCP Tool Integration: Full support for Model Context Protocol servers and tools
• Cross-platform: Native desktop apps for Windows, macOS, and Linux
• Advanced UI: Modern React interface with streaming responses and tool call visualization
• Flexible Configuration: Support for remote and local MCP servers
• Developer Friendly: Built with latest MCP SDK v1.17.2 for maximum compatibility

Quick Start

Prerequisites

• Node.js (v18+)
• pnpm package manager
• Ollama account with API access

Installation & Setup

1. Clone the repository:

   git clone <repository-url>
   cd ollama-turbo-coder

2. Install dependencies:

   pnpm install

3. Get your Ollama API key:

   • Sign up for an Ollama account
   • Create an API key at https://ollama.com/settings/keys
   • Copy the API key for the next step

4. Configure environment:

   # Create .env file
   echo "OLLAMA_API_KEY=your_api_key_here" > .env
   echo "OLLAMA_DEFAULT_MODEL=gpt-oss:120b" >> .env

5. Start the development server:

   pnpm dev

The app will launch automatically. Add your API key in Settings if you haven't set it in the .env file.

Available Models

Ollama Turbo currently supports:

• gpt-oss:20b - 20 billion parameter model
• gpt-oss:120b - 120 billion parameter model

Both models support:

• 131K token context window
• MCP tool integration
• Fast inference through Turbo acceleration

Using Ollama Turbo

The app connects to Ollama's Turbo service at https://ollama.com. Simply:

1. Launch the app
2. Add your Ollama API key in Settings
3. Select your preferred model (gpt-oss:20b or gpt-oss:120b)
4. Start chatting!

MCP Server Support

Ollama Turbo Desktop includes full MCP (Model Context Protocol) support with the latest SDK v1.17.2, enabling seamless integration with tools and external data sources.

Supported MCP Server Types

1. Remote MCP Servers

{
  "transport": "stdio",
  "command": "npx", 
  "args": ["-y", "mcp-remote", "https://mcp.buildaipod.com/mcp"]
}

2. Official MCP Servers

// Filesystem access
{
  "transport": "stdio",
  "command": "npx",
  "args": ["@modelcontextprotocol/server-filesystem", "/path/to/directory"]
}

// Git operations  
{
  "transport": "stdio",
  "command": "npx",
  "args": ["@modelcontextprotocol/server-git"]
}

// Web search via Brave
{
  "transport": "stdio", 
  "command": "npx",
  "args": ["@modelcontextprotocol/server-brave-search"]
}

3. Custom MCP Servers

// Python server
{
  "transport": "stdio",
  "command": "python3",
  "args": ["/path/to/your/mcp_server.py"] 
}

// Node.js server
{
  "transport": "stdio",
  "command": "node",
  "args": ["/path/to/your/mcp-server.js"]
}

Adding MCP Servers

1. Go to Settings → MCP Servers
2. Click Add Server
3. Configure your server details
4. Tools will be automatically discovered and available to AI models

MCP Features

• ✅ Auto-discovery: Tools are automatically detected and made available
• ✅ Health monitoring: Automatic reconnection and health checks
• ✅ Multiple transports: stdio, SSE, and StreamableHTTP support
• ✅ OAuth authentication: Full OAuth 2.1 support for authenticated servers
• ✅ Streaming responses: Real-time tool execution with streaming results

Building for Production

To build the app for your platform:

# For macOS
pnpm dist:mac

# For Windows
pnpm dist:win

# For Linux
pnpm dist:linux

Troubleshooting

API Key Issues

• Make sure you have created an API key at https://ollama.com/settings/keys
• Verify the API key is correctly entered in Settings

Connection Issues

• Check your internet connection
• Verify that https://ollama.com is accessible
• Try using the default API endpoint without custom URLs

Model Selection

• Only gpt-oss:20b and gpt-oss:120b are available in Turbo
• These models don't support image inputs

Technical Architecture

Built with modern technologies for optimal performance and maintainability:

Core Technologies

• Electron: Cross-platform desktop framework
• React 19: Modern UI with hooks and concurrent features
• Vite: Fast build tool and dev server
• Ollama SDK: Official JavaScript SDK for Ollama API integration
• MCP SDK v1.17.2: Latest Model Context Protocol implementation

Key Components

• ollamaHandler.js: Manages API communication and streaming responses
• mcpManager.js: Handles MCP server connections and tool discovery
• toolHandler.js: Processes tool execution with robust argument parsing
• authManager.js: OAuth 2.1 authentication for MCP servers

Recent Updates

• ✅ MCP SDK Updated: Upgraded from v1.7.0 to v1.17.2 for latest features
• ✅ Tool Compatibility: Fixed argument parsing for both object and string formats
• ✅ Conversation History: Proper tool call format handling in chat history
• ✅ Enhanced Security: OAuth 2.1 compliance with Resource Indicators support

Development Features

• Hot reload in development
• Comprehensive error handling and logging
• Automatic MCP server health monitoring
• Cross-platform command resolution

License

MIT

Acknowledgements

This project is based on Groq Desktop Beta and has been adapted to work with Ollama and other local LLM providers.