MCP publishes a new roadmap: five priority areas, named maintainers, and expedited review for SEPs that fit

The Model Context Protocol project published an updated roadmap on 2026-08-22, announced on the official blog by lead maintainers David Soria Parra and Den Delimarsky. It replaces the roadmap published in March 2026 and covers the next specification release and beyond. The structure is new: each of five priority areas names the Core Maintainers responsible for it and the Working Groups carrying the work, and Specification Enhancement Proposals that fall inside a priority area now receive expedited review. Proposals outside them are not rejected automatically, but the page is direct about the trade: maintainer review time goes to the priority areas first.

The first two areas extend the 2026-07-28 architecture. Agentic messaging primitives covers server-initiated events, channels and webhooks through the Triggers and Events Working Group, so clients are not left polling for the results of long-running work, plus a composition review to make Tasks, subscriptions/listen and progress notifications share one lifecycle, and continued work on the Tasks extension (SEP-2663) toward eventual inclusion in the core protocol. HTTP-native transport unification proposes Streamable HTTP as the single binding, spoken over stdin/stdout for local servers, which the page suggests could run as HTTP/2 over stdio, along with extending the revision's caching model (SEP-2549) toward ETags.

The third area, agent identity and enterprise-ready security, targets callers that are not people in browsers: finalizing Demonstrating Proof of Possession (DPoP) and defining an agent identity and delegation path built on Workload Identity Federation (SEP-1933), the ID-JAG grant behind Enterprise-Managed Authorization, and RFC 8693 token exchange, coordinated with the IETF OAuth and WIMSE working groups. An Agent Identity Working Group is forming to carry it. The fourth, improved primitives, commits to redesigning the tools/call result shape, because a response can carry the same output as both content and structuredContent and implementations have diverged, and starts a progressive discovery effort so servers with large catalogs can expose a small entry point instead of a hundred tools up front. A Core Primitives Working Group is forming for both.

The fifth area covers the SDKs, including an experiment in generating a candidate Tier 1 SDK and its quickstart examples from the specification and validating them against the conformance test suite. Looking back, the post credits the March roadmap's four areas with most of what landed in the 2026-07-28 release: the stateless core (SEP-2575, SEP-2567), server/discover, cacheable list results, the Multi Round-Trip Requests pattern (SEP-2322), Tasks moving to an official extension, and the authorization hardening that arrived as issuer validation, issuer-bound client credentials and Client ID Metadata Documents.