Model Context Protocol Locks Its Largest-Ever Spec Revision, Rebuilding the Core as a Stateless Protocol Ahead of a July Release
The 2026-07-28 MCP release candidate makes the protocol core stateless, formalizes an Extensions framework, and deprecates Roots, Sampling, and Logging. The RC locked May 21 for a July 28 final.
Overview
The Model Context Protocol, the open standard for connecting AI applications to external data and tools, has locked the release candidate for its next specification. According to the Model Context Protocol Blog, the new 2026-07-28 revision “is the largest revision of the protocol since launch and delivers on the 2026 roadmap.” Its headline change is a stateless protocol core, alongside a first-class Extensions framework, a redesigned Tasks capability, server-rendered user interfaces through MCP Apps, hardened authorization, and a formal deprecation policy.
The candidate was locked on May 21, 2026, with the final specification scheduled for publication on July 28, 2026, the version string it carries. That leaves roughly a ten-week window for implementers to validate against the candidate, and the Model Context Protocol Blog notes that Tier 1 SDKs are expected to ship support within it.
The revision arrives under new governance. As previously reported, MCP was contributed to the Linux Foundation’s Agentic AI Foundation, a directed fund the Agentic AI Foundation lists as co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, Amazon Web Services, Cloudflare, and Bloomberg. MCP joined the foundation alongside Block’s goose and OpenAI’s AGENTS.md.
What We Know
A stateless core
The central change is the move to a stateless protocol core. The previous 2025-11-25 specification assumed a persistent session between client and server, which is awkward to scale across distributed, multitenant deployments. Under the new model, according to the Model Context Protocol Blog, “A remote MCP server that previously needed sticky sessions, a shared session store, and deep packet inspection at the gateway can now run behind a plain round-robin load balancer, route traffic on an Mcp-Method header.”
Dropping persistent sessions creates a problem for any flow where the server needs to ask the client for something mid-call, such as an elicitation prompt. The specification narrows when that can happen: per the Model Context Protocol Blog, “Server-initiated requests may now only be issued while the server is actively processing a client request.”
Extensions, Tasks, and MCP Apps
The Extensions framework gains first-class status, letting capabilities ship on their own timeline rather than waiting on the core specification. Tasks, which shipped as an experimental core feature in 2025-11-25, has graduated into an extension. The Model Context Protocol Blog describes the redesign: a server can answer a tools/call with a task handle, and the client drives it with tasks/get, tasks/update, and tasks/cancel. Task creation is server-directed: the client advertises the extension and the server decides when a call should run as a task. Per the same source, “The Tasks extension reshapes the lifecycle around the stateless model.”
MCP Apps is also delivered as an extension. According to the Model Context Protocol Blog, “MCP Apps lets servers ship interactive HTML interfaces that hosts render in a sandboxed iframe.”
Authorization and deprecations
Six SEPs harden the authorization specification to align more closely with how OAuth 2.0 and OpenID Connect are deployed, according to the Model Context Protocol Blog. One of them, SEP-2468, requires clients to validate the iss parameter on authorization responses per RFC 9207 as a mitigation for mix-up attacks.
The candidate also brings a formal deprecation policy and exercises it: Roots, Sampling, and Logging are deprecated. The Model Context Protocol Blog points implementers to replacements, citing tool parameters, resource URIs, or server configuration in place of Roots; direct integration with LLM provider APIs in place of Sampling; and stderr for stdio transports plus OpenTelemetry for structured observability in place of Logging.
What We Don’t Know
The specification is a release candidate, not a final standard; the Model Context Protocol Blog frames the period through July 28 as a validation window, and details could still shift before publication. The candidate also does not name specific quotes from its two listed maintainers, David Soria Parra and Den Delimarsky, beyond the technical description, so the motivation behind individual design choices is documented through the SEPs rather than direct commentary.
Analysis
The practical thrust of the revision is operational rather than feature-driven. By making the core stateless, MCP trades the convenience of persistent sessions for deployment patterns that fit commodity HTTP infrastructure, the kind of round-robin, load-balanced setup that large providers already run for everything else. The Extensions framework is the structural counterpart: by pushing Tasks and MCP Apps out of the core and onto independent release cadences, the protocol can keep its stable surface small while still evolving the riskier, faster-moving pieces. For a standard that, by the Agentic AI Foundation’s own description, has become infrastructure for integrating LLM applications with external data and tools, that separation is a bet on longevity under neutral governance rather than on any single new capability.