Stephen Downes

This article offers a short explanation of what the model context protocol (MCP) is exactly and then outlines a number of ways it can go wrong. I'm not going to disagree with any of these criticisms except to observe that they are all basically versions of "MCP as it currently stands is a big security hold". That is to be expected with you give an AI generally unconstrained access to your data and services. To make a long story short: MCP is obviously an idea still in development, and this article offers a rough description of the sort of infrastructure that will need to be in place before we can depend on it in production.