Running MCP 38 Serverless TextGen Hub ♨ 38 Now supports inference providers and multimodal. No GPU req.