Lemonade, from Lemonade Inc., is an MCP server tool providing local AI capabilities. It integrates local models for various tasks, connecting with many MCP-compatible clients like GitHub Copilot and Claude Desktop. The program also supports applications using OpenAI, Anthropic, and Ollama API standards. Developed with C++, Rust, Python, and React, Lemonade is distributed via native installers, package managers, Docker images, and an embeddable binary. It offers private AI processing without external service charges, optimized for specific hardware but supporting NVIDIA GPUs and x86 CPUs.

What does Lemonade connect to?

Lemonade functions as a Model Context Protocol (MCP) server, enabling local AI and LLM operations. It connects with any MCP-compatible client, including GitHub Copilot, Claude Desktop, MCP Inspector, Cursor, the mcp Python client, Claude Code, and Cline. Beyond MCP, Lemonade extends its reach by supporting connections from applications that utilize OpenAI, Anthropic, and Ollama API standards. This broad compatibility allows it to serve as a central hub for local AI model inference across a diverse ecosystem of client applications.

Core features and tools

Lemonade exposes its capabilities through a set of core MCP methods and specific callable tools. The core methods include initialize for protocol negotiation, notifications/initialized for client acknowledgment, tools/list to catalog callable tools with JSON Schemas, tools/call to invoke specific tools, and ping for liveness checks. Specific tools callable via tools/call are lemonade_list_models for model discovery, lemonade_chat for chat completion, lemonade_transcribe_audio for audio transcription, lemonade_generate_image for image generation, and lemonade_omni for multimodal interactions.

Who is it for?

Lemonade is for users and builders keen on running local AI, developers, and LLM enthusiasts. It is optimized for PCs featuring AMD Ryzen AI, Radeon, and Strix Halo hardware, yet it also supports NVIDIA GPUs and x86 CPUs. This program delivers private local AI capabilities. Skip this if your primary need is strictly cloud-based AI services without any local processing requirements or interest in integrating local models.

How to get started

The server authenticates using an API key passed as an environment variable. All API routes enforce authentication when LEMONADE_API_KEY is set, and Lemonade supports the MCP Streamable HTTP transport protocol.

Use cases

• Running local AI models: Manage models for chat, coding, speech, and image generation via GUI/CLI or third-party OpenAI, Anthropic, or Ollama-compatible applications.
• Building AI agents: Use Lemonade as an MCP server to expose local models as tools, allowing LLMs to interact with external services for Tiny Agents.
• Self-hosting AI: Install as a lightweight local service, secured and accessible across a local area network for personal or household use.

Lemonade offers a versatile platform for local AI, connecting diverse clients and supporting multiple API standards. It provides private AI processing on various hardware, including specialized AMD components, NVIDIA GPUs, and x86 CPUs. The tool-calling mechanism enables advanced agent development and self-hosted model deployment.