"wgpu" matching MCP servers:

An MCP stdio server that lets an AI agent run ComfyUI or VoiceStudio batch jobs on a GPU shared with a local LLM server by unloading the LLM, launching the executor, running the jobs, and always reloading the LLM afterward. It exposes comfy_batch, voice_batch, and gate_status tools so a single blocking call handles the whole GPU hand-off safely even if the client disconnects.

Let me evaluate the README content.