Open MCPs as the New Web Schema is Brian Swichkow's thesis that open implementations of the 📝Model Context Protocol (MCP) are becoming the web's interface for AI agents, and that their openness bears on who controls access to human knowledge.
In Brian Swichkow's words, "Google is forcing the web to go in a direction where MCP is going to be the new schema for the web." The Google claim is his reading of the web's direction, which he calls "the web starting to rearchitect itself so that it's more easily digestible by agents." On that reading, a network built for humans browsing websites becomes one where agents access context directly, retrieve structured knowledge, and operate across applications rather than through isolated APIs or human interfaces.
The word "open" carries the argument. An open MCP is standardized, interoperable, portable, and not locked into a single provider. The concern is centralized control: whoever controls contextual memory and AI access layers may come to control much of human knowledge interaction, and closed implementations would carry platform lock-in and extractive platform dynamics into the agent era. The position extends Brian's argument in 📝Sovereign Data, Not Just Owned Data.
Brian also argues that models are becoming commoditized and that lasting value lies in context, memory, and structured knowledge, with MCPs as the infrastructure layer beneath them. 📝Connectivity Is the New Moat makes a related case, and 📝Public MCP Servers covers how a website joins this layer. Brian explains what an MCP does through the 📝Fire Control Panel for AI metaphor.
I use the word open on purpose. If models become commodities, value moves to context and memory, and whoever holds that layer holds how people meet knowledge. I'm building 📝MythOS as one such layer, and 📝MythOS MCP is how agents reach it.