Anthropic Releases MCP 2026-07-28 With Stateless Core Architecture
Key Takeaways
- ▸Stateless core architecture eliminates session management, enabling MCP servers to run natively on serverless platforms and edge infrastructure
- ▸Standardized extensions framework (MCP Apps and Tasks) provides a formal path for adding interactive UIs and long-running work capabilities without modifying the core protocol
- ▸Enterprise-grade authorization now aligns with OAuth 2.0 and OIDC standards, integrating directly with production identity systems like Microsoft Entra and Okta
Summary
Anthropic has released MCP (Model Context Protocol) version 2026-07-28, marking one of the most significant updates to the industry-standard protocol for connecting AI agents to applications and data. The new spec introduces a stateless core architecture, moving from a bidirectional stateful model to a request/response design that enables deployment on serverless and edge infrastructure—simplifying server development and scaling.
MCP has experienced explosive growth, surpassing 400 million monthly SDK downloads (a 4x increase in 2026) and establishing itself as the de facto standard for agentic integration. The 2026-07-28 release adds formalized extensions through the MCP Apps and Tasks frameworks, and hardens authorization with OAuth 2.0 and OIDC compliance for enterprise deployments on identity platforms like Entra and Okta.
Support for the new spec is being rolled out across Claude products, with over 950 MCP servers now listed in Anthropic's connectors directory. Major ecosystem partners—including Figma, Intuit, Netlify, Zoom, and others—are actively adopting the spec, citing faster deployment, improved scalability, and stronger enterprise integration as key benefits.
- Protocol has reached critical mass: 400M+ monthly SDK downloads (4x YoY growth) with 950+ servers listed and adoption from major companies across infrastructure, design, finance, and collaboration platforms
- Stateless design simplifies developer experience and reduces operational complexity, allowing companies to deploy and scale MCP servers faster across their platforms



