The open-source Lemonade AI server for "100% free and private" AI usage across Windows and Linux in leveraging AMD Ryzen AI NPUs, Radeon GPUs, and x86_64 CPUs, is now much more powerful with today's v10.8 release.

The AMD-backed Lemonade local AI server continues rapidly advancing and with today's v10.8 release it adds Model Context Protocol (MCP) server integration. With Lemonade's MCP server support, any MCP-compatible client can interact with the Lemonade server in treating your locally-running models as tools. GitHub Copilot, Claude Desktop, Cursor, and other MCP-supportive clients can now interact with Lemonade via MCP with capabilities like chat, audio transcription, image generation, and one-shot multi-modal handling with Lemonade Omni.

All the new MCP server capabilities are outlined via the new documentation in Lemonade 10.8.

"Why it matters. Agents running in a frontier model can now route the privacy-sensitive or high-volume parts of a task to local Lemonade models without leaving the conversation. Bulk classification, on-device transcription and image generation become free, private and offline, orchestrated by whatever host model the user prefers."

More details in this pull request adding the MCP server support.

Lemonade 10.8 also now enables ROCm for the GFX1152 / Radeon 860M graphics, adds support for live model management, Moonshine speech-to-text, and even experimental NVIDIA GB10 ARM64 support using the Llama.cpp CUDA back-end.

Downloads and more details on Lemonade 10.8 via GitHub.