The Agentic AI Foundation Updates Core Features of the Model Context Protocol
The Agentic AI Foundation (AAIF) is announcing a major update to Model Context Protocol (MCP) that represents the largest revision of the protocol since its initial launch—removing core scaling bottlenecks, finalizing the project's transition to a stateless architecture, and establishing the formal governance and security hardening required for enterprise-scale deployments.
Historically, running MCP at scale required sticky routing or shared state to maintain continuity across sessions. This made large-scale production deployments complex to implement and operate even when the capabilities being exposed were stateless.
According to the foundation, using feedback from the community of builders creating MCP Hosts & Clients, the Core maintainers met in December 2025 to lay out the foundational changes the protocol needed in the Future of MCP Transports.
This update officially finalizes MCP's transition to a completely stateless architecture. By shifting to a stateless model, organizations can now run MCP servers behind standard load balancers using the existing Kubernetes and cloud-native DevOps tools they already rely on.
This completely removes the operational friction of managing session state, allowing teams to treat MCP servers like ordinary, highly scalable HTTP services, the foundation said.
Newest features and capabilities include:
- Stateless architecture that scales on HTTP infrastructure
- Enterprise-grade security that aligns with the security standards companies already use (OAuth 2.0 and OpenID Connect)
- Formal governance via a predictable, 12-month deprecation policy
- New extensions like MCP Apps so developers can build richer AI apps with interactive interfaces and long-running tasks
To bring mature, internet-standard predictability to the ecosystem, this release introduces a Formal Deprecation Policy alongside a new Extensions framework.
Under the new policy, developers are guaranteed a minimum window of twelve months between a feature's formal deprecation and its earliest possible removal. This ensures that enterprise engineering teams can build against the MCP specification with total confidence.
Because of MCP's typical single-client, many-server deployment pattern, it faces unique security vectors and policy enforcement challenges for security teams, the foundation said.
This release includes significant authorization hardening changes to address security issues by aligning the MCP auth spec with how OAuth 2.0 and OpenID Connect are deployed in practice.
Crucially, the protocol now enforces the mandatory validation of the iss parameter. This protocol-level defense closes an entire class of mix-up attacks, ensuring secure identity and context delegation across distributed environments.
Additionally the Enterprise Managed Authorization allows organizations to centrally manage authorization for MCP Servers, and simplifies the login experience for end-users by making the organization's IdP the authoritative decision-maker for MCP server access.
In addition to core architectural and security improvements, the update expands what AI agents can actually deliver back to users. MCP Apps can now ship their own server-rendered, interactive interfaces.
MCP Apps is an official protocol extension, which allows it to evolve independently of the core specification release cycle. MCP Apps allows developers to deploy rich, interactive user experiences natively from their own servers. This dramatically accelerates the development lifecycle for feature-rich, user-facing agentic applications.
Not every tool call finishes in a single round trip. Expensive computations, batch jobs, and other long-running tool calls don't fit the request-response model. MCP Tasks let servers return a durable handle instead of blocking, so clients can poll for progress, provide input when needed, and retrieve the result after reconnecting.
MCP Tasks has graduated from an experimental feature to an official extension to provide a common way of supporting async long running operations.
“Since being contributed to the AAIF under the Linux Foundation in December 2025, MCP's mission has been to establish open, vendor-neutral standards for AI agent communication. This latest update represents the critical bridge between experimental agentic loops and production infrastructure,” the foundation said.
For more information about this news, visit https://aaif.io.