July 29, 2026, (Inside AI) — The Model Context Protocol (MCP) has released its fifth specification, MCP 2026-07-28, delivering a fundamental architectural shift to a stateless core while strengthening enterprise security and graduating official extensions. Anthropic confirmed the update is rolling out across Claude products.
The protocol, which has become the industry standard for connecting AI agents to applications, surpassed 400 million monthly SDK downloads this year, a 4x increase. The new spec marks one of the most significant releases since MCP's inception, designed to simplify server deployment and scale adoption.
At the heart of the update is the move from a bidirectional stateful protocol to a request/response model. This stateless core means servers can now deploy on serverless and edge infrastructure, a change that directly addresses developer friction in building and scaling MCP servers for Claude and other agents.
Anthropic also introduced a versioned extensions framework, standardizing MCP Apps and Tasks. This gives developers a formal path to add capabilities like interactive UIs and long-running work without altering the core protocol. Authorization has been hardened to align with production OAuth 2.0 and OIDC deployments, enabling MCP servers to connect to enterprise identity systems like Entra or Okta without workarounds.
Companies across the ecosystem have been building on the new spec alongside the MCP community since beta. The shift to statelessness resolves a long-standing tension between MCP's original design and modern cloud-native architectures. Early MCP versions assumed persistent connections, which complicated deployment on platforms like AWS Lambda or Cloudflare Workers. Now, each request carries full context, making servers horizontally scalable and easier to manage in production.
This evolution mirrors broader industry trends. Protocols like HTTP succeeded because of statelessness, and GraphQL subscriptions later reintroduced state selectively. MCP's new extensions framework borrows from this pattern, keeping the core simple while allowing opt-in complexity. The OAuth 2.0 alignment is critical for regulated industries, where security audits often block non-standard protocols. Anthropic's move reduces that barrier, potentially accelerating MCP adoption in finance and healthcare.
However, statelessness introduces trade-offs. Real-time collaboration or streaming use cases may require additional infrastructure, like message queues or WebSocket bridges, which the Tasks extension might address. The spec does not detail latency implications for high-frequency agent interactions. Competitors like Google's A2A protocol have emphasized streaming-first designs, and it remains to be seen whether MCP's request/response model can match that performance for live agent-to-agent communication.
The MCP community has grown rapidly since Anthropic open-sourced the protocol in 2024. Its rise parallels the agentic AI boom, where models like Claude, GPT-4, and Gemini need a common language to interact with tools, databases, and each other. MCP's stateless core may also simplify compliance with data residency laws, as servers can process requests without storing session data across regions.
Anthropic has not disclosed a timeline for deprecating older MCP versions, but the company's rollout across Claude products signals a strong commitment. Developers can review the full specification on the official MCP spec site.
