Running MCP 31 31 Serverless TextGen Hub ♨ Now supports inference providers and multimodal. No GPU req.