The MCP 2026-07-28 specification has done something unusual in the world of developer tools: it has made a deliberate choice to remove the initialize handshake and session header, replacing them with required method and tool-name headers. The stated goal is practical. Gateways routing agent traffic should not have to parse JSON to know where a request is headed. That is a real efficiency gain, and it is also a philosophical statement. The reaction has split between developers who see this as a rediscovery of REST and those who argue the standard itself was always the point. Both sides are right, and that is what makes this worth paying attention to.
For anyone who has watched the Navigating AI/ML Job Requirements: A Shift in Expected Skills conversation, this feels familiar. The industry keeps circling back to the same tension between simplicity and capability. Stateless protocols are easier to scale, easier to debug, and easier to reason about. The trade-off is that you lose the session context that made stateful interactions feel natural. What the MCP maintainers have done is acknowledge that most agent traffic does not need a persistent session. It needs a clear method, a tool name, and a payload. That is closer to how Scale AWS Server Deployments Effortlessly with Stateless Model Context Protocol describes the operational benefits: no sticky sessions, no session affinity, just requests that any gateway can route without holding state. That is not a regression. It is a refinement.
The developers calling this a rediscovery of REST are not wrong. A stateless protocol with explicit headers is how HTTP was designed to work. But the more interesting point is what the standard itself represents. MCP was never just about the handshake. It was about giving agents a common vocabulary for tools, methods, and responses. Stripping away the session layer does not weaken that vocabulary. It strengthens it by making the protocol more transparent. The method and tool-name headers are not a concession to legacy thinking. They are an admission that routing should be a first-class problem, not an afterthought solved by parsing bodies. That is a mature design decision, and it deserves more credit than the "it is just REST" dismissal gives it.
What we would tell a reader who asks about this is straightforward. If you are building agent infrastructure, the stateless shift is a win. It removes a class of operational friction without sacrificing the semantic clarity that makes MCP useful. The developers who are upset are likely the ones who built tooling around the session model, and their concern is legitimate in the short term. But standards evolve, and the direction here is toward lower friction and broader compatibility. The open question to watch is how the ecosystem handles the transition. Will gateway vendors adopt the new headers quickly, or will there be a period of fragmentation? That is the detail that matters. The protocol is cleaner now, but a cleaner protocol only helps if the ecosystem moves with it. We would bet on that movement, because the alternative is a return to the JSON-parsing chaos this change was designed to eliminate.