The headline of the new MCP roadmap is the move toward a stateless core, but the more telling signal is what the five priority areas say about where the protocol now thinks the hard problems live. They’ve shifted from “how do I connect a tool” to “how do I operate agents I can trust in production.”
- 🎯 Agentic messaging primitives — server-initiated events, webhooks, and a maturing Tasks extension. Request-response was never going to carry long-running, mid-flight-steerable agent work.
- ⚡ HTTP-native transport — local servers speaking Streamable HTTP over stdio. Collapsing two deployment models into one is the boring kind of win that deletes a whole class of bugs.
- 🔍 Agent identity and enterprise security — DPoP and Workload Identity Federation instead of pasted API keys and long-lived tokens. This is the line item that actually unblocks enterprise deployment.
- 💡 Improved primitives — progressive discovery, so a server exposes a small entry point and reveals more of its catalog as the conversation narrows. Anyone doing context engineering will recognize this immediately.
- 📊 SDK developer experience — conformance testing and docs, aimed squarely at the fact that people now build MCP servers using agents.
Organizing the roadmap by priority area instead of dates is itself the message: MCP is governed by Working Groups now, not one vendor’s release train. The HN discussion splits about how you’d expect — relief at a stateless core on one side, groans from teams who built around the stateful model on the other.
The wider read is broadly positive but not uncritical. Obot treats the roadmap as MCP graduating into production-grade, enterprise-ready infrastructure, and GetKnit sees it cementing MCP as the universal integration standard. The Register is more measured, calling the stateless shift smart and overdue for scalability while warning that anyone running a custom stateful implementation faces real migration uplift. That tension — a cleaner protocol against the cost of getting there — is the thing to watch as the next spec lands.
tags: [ agentic-ai ] [ ai-infrastructure ] [ enterprise-ai ]