The Linux Foundation’s Agentic AI Foundation announced on July 28, 2026 an updated Model Context Protocol (MCP) designed to make AI automation easier for enterprises. MCP, originally open sourced by Anthropic in November 2024, defines a client‑server interface that lets AI agents such as those based on GPT-5.6 Sol or Claude Opus 5 connect to data sources, tools, or other applications. David Soria Parra, a member of technical staff at Anthropic and co‑inventor of MCP, said in a blog post that the release is the most important since the first remote MCP launch and represents a leap in serving scalable MCP servers. Caitie McCaffrey, a Microsoft software engineer and core MCP maintainer, explained that removing session state reduces complexity for large‑scale deployments and enables standard load balancing behind Kubernetes and other DevOps platforms. The new specification eliminates the legacy stateful architecture, allowing MCP servers to run behind standard load balancers on existing Kubernetes tooling. Version 2026‑07‑28 also introduces a Specification Feature Lifecycle and a Deprecation Policy that mandates a minimum 12‑month interval between deprecation and removal, providing enterprises with a predictable migration timeline. Security enhancements include Specification Enhancement Proposal 2468, which requires an issuer (iss) parameter in authorization responses to prevent OAuth Mixup attacks, and support for the Enterprise Managed Authorization extension, enabling management through a central identity provider. The update adds an async task extension that turns long‑running or batch operations into durable, non‑blocking requests, allowing task identifiers to be stored for resumption after failures. Additional capabilities include header‑based routing and cacheable list results, though developers who rely on session identifiers may need to adjust their code.