Stateless MCP and the moment agent plumbing became infrastructure

The most consequential agent news of the cycle is a protocol revision that removed a handshake. Industrialisation is never glamorous — it is the moment a craft technique becomes something you no longer think about.

The Model Context Protocol completed its largest rewrite since launch, removing protocol-level sessions and the initialize handshake so any server replica can handle any request. Translated out of the spec language: agent backends can now scale like ordinary web services instead of pinning each conversation to a specific server. That is the difference between a demo and a fleet.

Why removing a handshake is the whole story

Session affinity — the requirement that every message return to the server holding its state — is what made agent fleets expensive, because it defeats the load balancers and autoscalers that make ordinary web traffic cheap. Self-contained, stateless requests can be routed, cached, and traced by any replica. The change is invisible to a user and decisive to an operator, which is exactly what industrialisation looks like.

It matters now because of scale. MCP SDK downloads reached roughly 97 million a month, and a protocol at that volume cannot carry an architecture that fights horizontal scaling. The revision arriving under the Linux Foundation's shared stewardship — with OpenAI, Google, Microsoft, AWS, Salesforce, and Snowflake behind it — is what makes it an operator's wishlist rather than one vendor's preference.

But cheaper plumbing does not clear the real wall

Here is the tension the cycle exposes. Only 11–14% of enterprise agent pilots reach production, and the majority stall on identity, audit, and access control — not capability. Stateless MCP makes fleets cheaper to run, but a cheap-to-run agent still hits the same governance wall. The bottleneck has moved from "can it reason" to "can we let it act", and no protocol revision answers that.

So the plumbing standardises while the deployments stall, and both are true at once. The lesson is that the next wave of agent products competes on governance — identity, scoped authority, audit trails — not on intelligence. The infrastructure is nearly solved. The accountability is not.

MLQ.ai — MCP's stateless rewrite targets enterprise-scale AI agents → · The Agentics — The enterprise MCP guide 2026 →