One MCP Server for Six AWS Inference Backends: Bedrock, SageMaker, GPUs, Inferentia and Trainium

A step by step look at the MCP server that runs a six-backend Gemma 4 demo on AWS: one Python file over the demo's own code, eight tools marked read-only, write or destructive, SSM Run Command instead of open ports, and an agent that counts through the server and gets the same grade as in-process tools.