Making MCP stateless is the least exciting and most consequential change of the quarter
Protocols become infrastructure when they stop being special. A stateless MCP can sit behind an ordinary load balancer, scale horizontally and be operated by people who know nothing about agents — which is the precondition for it being deployed by anyone other than enthusiasts.
The MCP 2026-07-28 release candidate moves session state into explicit handles, capabilities into negotiated extensions, and sharpens authorization rules. None of that is a feature anyone will demo. All of it is what a protocol needs before a platform team will take it.
Statefulness was the deployment blocker
A stateful protocol requires sticky sessions. Sticky sessions require a load balancer that understands the protocol. That requirement meant MCP servers could not be deployed the way every other service in an estate is deployed — and anything that cannot be operated with existing tooling gets operated badly or not at all.
Explicit handles fix that by making the state something the client carries rather than something the server remembers. It is an old pattern, and it is old because it works.
Negotiated extensions are the survival mechanism
A fixed capability set forces a specification revision every time a vendor needs something. That is how protocols ossify: the committee becomes the bottleneck, vendors ship proprietary side channels, and the standard becomes a description of the least interesting common subset. Negotiation lets the edges move without the centre having to approve.
What arrives next, and what it breaks
Vendors are already shipping endpoints as a matter of course — MESCIUS published one for its developer tools on 30 July, which is unremarkable in itself and precisely why it matters. Once every vendor ships one, an agent's effective capability becomes the union of the servers it is pointed at, and nobody owns that set.
That is the governance problem the orchestration products are quietly forming around. Framing the unit of work as task-to-pull-request puts agent output inside a review boundary that already exists. Sharper authorization in the protocol plus a human-reviewed output surface is a defensible posture. Neither alone is.
The agent industry spent two years on capability and is now, belatedly, building the plumbing. Plumbing is what decides whether any of it ships.
Agentic AI Foundation — MCP Is Growing Up → · Agentic.ai — Agentic AI News — July 2026 Launches, Models & Research →