MCP roadmap: AI agent support and HTTP unification next

The roadmap comes as MCP has solidified its role as the de facto standard for AI connectivity, transitioning from a single-company initiative to a Linux Foundation project. The focus on agentic messaging primitives and server-initiated events reflects the technological shift from simple chat interactions to long-running autonomous tasks. This change is driven by the expectation that AI agents will act independently with their own identities, delegating permissions to sub-agents, which requires a more robust and secure authentication model than the current human-in-the-browser approval process.

The push to unify transport on HTTP is a practical step to simplify and strengthen the protocol, leveraging the 2026-07-28 spec that already introduced HTTP support for remote servers. Similarly, improving primitives and SDK developer experience aims to reduce complexity for both human developers and AI tools that increasingly implement MCP. These efforts collectively point toward a more scalable, enterprise-ready protocol that can handle the demands of production-grade agentic workflows, though the roadmap outlines directions rather than concrete timelines.