The Model Context Protocol, a system for connecting AI models to applications, has released a major specification update designed to make enterprise deployment simpler and more stable.
The new spec includes a formal 12-month deprecation policy—giving companies time to plan for changes—and represents a fundamental redesign from its original local-machine-only foundation to support large-scale corporate use.
What happened
The Model Context Protocol (MCP)—a system that connects AI models to applications and services—has released a new specification redesigned for enterprise deployment. The update includes a formal deprecation policy guaranteeing at least 12 months between when a feature is marked for removal and its actual removal, except for critical security fixes.
Why it matters
MCP originally ran only on local machines; the new spec rethinks its foundations to work at enterprise scale, removing a major barrier to wider corporate adoption. Companies using MCP—including those relying on it to integrate AI into business workflows—now have stability guarantees when planning upgrades.
What to watch
MCP is managed by the Agentic AI Foundation under the Linux Foundation. While Anthropic introduced it nearly two years ago and maintains significant influence, OpenAI, Google, Microsoft, and Amazon also contribute, and many developer tools and business software now support it.
The Model Context Protocol (MCP) has undergone a significant redesign to address what has been the main barrier to enterprise adoption. The new specification, documented in the protocol's official documentation, fundamentally rethinks MCP's architecture and policies to make large-scale corporate deployment practical.
Central to the update is a formal deprecation policy: any feature marked for deprecation must remain available and functional for at least 12 months before it can be removed from the protocol. The only exception is for critical security updates, which may be applied immediately. This change reflects a deliberate pivot toward enterprise requirements; large organizations need predictability and time to plan upgrades, and MCP's original model—which allowed faster iteration—was incompatible with corporate governance cycles.
The specification itself represents a major architectural rethink. MCP originated as a system that ran solely on local machines, connecting language models to applications and services on the user's device. The new spec fundamentally reimagines this foundation to work at enterprise scale, where systems must span multiple machines, environments, and organizational boundaries. While the documentation does not detail every technical change, the shift from local-only to distributed architecture is substantial.
MCP was introduced by Anthropic nearly two years ago and is still managed by the Agentic AI Foundation, which operates under the Linux Foundation. Although Anthropic remains influential—some principal maintainers currently work there—the protocol has grown into a genuine multi-vendor standard. OpenAI, Google, Microsoft, and Amazon all contribute to its development. Beyond the major cloud companies, MCP has gained traction with many developer tools and an expanding range of business software used for knowledge work and creative tasks. This breadth of adoption suggests MCP is becoming infrastructure rather than proprietary technology, though final authority rests with individual maintainers rather than any single company.
The Model Context Protocol addresses a growing need in enterprise AI: reliable, standardized ways for language models to interact with business software and services. MCP began as a local-machine-only system—suitable for individual developers or small teams connecting models to personal applications—but the new specification signals a maturation driven by demand from larger organizations. The 12-month deprecation window is a deliberate nod to enterprise reality: large companies cannot pivot quickly, and unpredictable breaking changes erode confidence in adopting emerging standards.
The governance structure is noteworthy. Although Anthropic originated MCP and some principal maintainers work there, the protocol now sits under the Linux Foundation's Agentic AI Foundation, with OpenAI, Google, Microsoft, and Amazon all contributing. This mix of competitors collaborating on a shared standard—reminiscent of how HTTP or TLS evolved—suggests the industry sees MCP not as a proprietary moat but as infrastructure. The inclusion of many developer tools and business software implies the protocol is moving from niche to mainstream use.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Google, Microsoft, and Meta continue expanding capital spending on AI infrastructure, signaling confidence tha…
ByteDance is reorganizing Doubao (its AI chatbot), Lark (workplace software), and Volcengine (cloud infrastruc…
Pegatron chairman Tzu-hsien Tung stated he remains optimistic about the long-term outlook for artificial intel…
Powertech Technology chairman DK Tsai said the company's FOPLP (Fan-Out Panel-Level Packaging) project with AM…
Amazon reported second-quarter 2026 net sales of US$200.61 billion（約32兆円） and net income of US$62.65 billion（約…
Nebius Group (NBIS) has more than doubled this year on strong demand for AI data centers with secured power, a…
The AI news that matters, in one minute each morning.
Sign up free