Anthropic Begins Claude Rollout of Stateless MCP Specification

Anthropic begins Claude rollout of the latest MCP specification

Anthropic says it is rolling out support for Model Context Protocol (MCP) specification 2026-07-28 across its Claude products. The announcement, published on Claude’s blog, describes the release as the fifth specification update and highlights changes to MCP’s core protocol, extensions framework, and authorization model.

MCP is an open protocol for connecting AI systems with external applications, tools, and data sources. Anthropic said MCP software development kits had surpassed 400 million monthly downloads, a fourfold increase during the year. That figure comes from Anthropic and was not independently verified in the supplied sources.

A stateless request-and-response core

The main technical change is a shift from a bidirectional, stateful protocol to a stateless request-and-response model. The official MCP 2026-07-28 specification describes the base protocol as using stateless, self-contained requests and capability negotiation on a per-request basis.

Under the new model, servers do not need to maintain persistent protocol session state between requests in the same way as earlier implementations. Anthropic says this can make MCP servers easier to deploy on serverless and edge infrastructure, while simplifying the process of building and scaling servers used with Claude.

Those benefits will still depend on each implementation’s approach to authentication, data storage, task execution, and other application-specific requirements. The Model Context Protocol Blog also identifies the stateless request-and-response core as the central feature of the release.

Versioned extensions for apps and tasks

The specification introduces a formal, versioned framework for extensions. Anthropic says MCP Apps and Tasks now use that framework rather than requiring changes to MCP’s core protocol.

MCP Apps are designed to support interactive user interfaces connected to MCP-enabled experiences. Tasks address work that can continue over a longer period. A separate extension path could allow these capabilities to evolve while keeping the base request-and-response layer relatively small and stable.

The official specification’s per-request capability negotiation is relevant to this design. Clients and servers can determine which optional capabilities are available for a particular interaction instead of assuming that every implementation supports the same features.

Authorization for enterprise identity providers

Anthropic also says the new release strengthens authorization by aligning MCP with production deployments of OAuth 2.0 and OpenID Connect, or OIDC. According to the company, this allows MCP servers to connect with enterprise identity providers such as Microsoft Entra and Okta without custom workarounds.

OAuth 2.0 is commonly used for delegated authorization, while OIDC adds an identity layer to the system. Support for those standards could help organizations apply centrally managed access controls when AI applications call internal services, although the exact setup will depend on individual deployments.

Anthropic said companies across the ecosystem worked with the MCP community on the specification during its beta period. The supplied announcement does not provide a separate implementation schedule for each Claude product; it says support is being rolled out.

For developers, the release provides a target based on a stateless protocol core, versioned optional capabilities, and standards-based authorization. The MCP specification remains the authoritative reference for implementation details, while Anthropic’s announcement describes the company’s Claude support.