This week, the Model Context Protocol (MCP), an open source standard for how AI systems interact with external tools and data sources, saw its largest update since its introduction. Most notably, MCP’s protocol core is now stateless, so requests are no longer dependent on a session tied to an individual server instance. This change has the potential to address long-standing barriers to scalability.
The blog post announcing the specification, written by lead maintainers David Soria Parra and Den Delimarsky (who both work at Anthropic), says:
“The highlight of this release is a stateless protocol core – MCP is transforming from a bidirectional stateful protocol into a request/response stateless protocol. It was one of the most highly-requested features from developers who were eager to get better reliability and scalability for their MCP servers.”
This update is “MCP’s most important since remote MCP first launched over a year ago,” Soria Parra wrote.
Why the Shift from Stateful to Stateless Matters#
In the initial releases of MCP, remote server communication relied heavily on maintaining bidirectional, stateful connections (such as persistent WebSockets or state-bound HTTP streams). While stateful connections excel at quick, full-duplex communication in localized settings, they create severe operational challenges when deployed in modern cloud platforms and enterprise distributed systems.
STATEFUL MCP (Legacy):
[ AI Client ] <---> (Sticky Session) <---> [ Specific MCP Server Instance ]
* Server failure loses state
* Hard to auto-scale horizontally
STATELESS MCP (New Spec):
[ AI Client ] ---> [ Standard Load Balancer ] ---> [ Any Stateless MCP Instance ]
* Infinite horizontal scale
* Simple container restartsThe Challenges of Stateful MCP#
- Sticky Sessions & Load Balancing Bottlenecks: Stateful connections force infrastructure layer routers to maintain sticky sessions. If a server instance experiences high load, traffic cannot be dynamically redistributed to idle instances without breaking the connection.
- Horizontal Scaling Friction: Auto-scaling groups (such as Kubernetes pods, container instances, or serverless workers) cannot easily spin down instances without dropping active client sessions and losing local context.
- Resilience and Server Restarts: If an individual MCP server crashes, any active session tied to that process is severed, forcing the orchestration client to re-initialize tool availability and state from scratch.
The Advantages of Stateless MCP#
By moving to a pure request/response stateless protocol core:
- True Elastic Scalability: MCP servers can now sit behind standard, off-the-shelf cloud load balancers. Requests are routed to any available instance in a pool without requiring state synchronization between nodes.
- Serverless Compatibility: MCP tool servers can now run efficiently on scale-to-zero serverless platforms (such as Azure Container Apps, AWS Lambda, or Cloudflare Workers) since instances do not need to stay running to hold open WebSockets.
- Fault Tolerance: If a container or worker node dies, the very next request is seamlessly handled by another node without client-side session destruction.
Technical Highlights of the New Specification#
Beyond the transition to a stateless core, this milestone release introduces several core architectural updates aimed at expanding enterprise capabilities and developer ergonomics:
1. Multi Round-Trip Requests#
Complex agentic workflows often require intermediate clarification, confirmation, or multi-step input gathering before completing an action. Multi round-trip request handling allows MCP servers and clients to negotiate multi-stage executions while keeping the underlying protocol stateless.
2. Header-Based Routing#
With state removed from the transport core, context propagation (such as correlation IDs, tenant identifiers, and routing metadata) is now handled natively via standard HTTP/gRPC headers. This enables modern service meshes and API gateways to route, filter, and inspect MCP traffic effortlessly.
3. Cacheable List Results#
Tool capabilities, schema definitions, and resource listings can now be cached using standard cache headers. This significantly decreases latency and eliminates unnecessary redundant network calls when LLMs evaluate available tool definitions.
4. Authorization Hardening#
Security models have been updated to align with stateless cloud design. Authorization tokens and security scopes are evaluated per request, preventing authorization drift and bringing MCP into compliance with modern Zero-Trust enterprise networks.
5. Formal Extensions Framework & Updated Tier 1 SDKs#
A formal extension model now allows custom domain capabilities to be built on top of the base protocol without risking core compatibility. In tandem, official Tier 1 SDKs have been updated to reflect these stateless patterns out of the box.
Enterprise Governance: Predictable Deprecation Policies#
Building enterprise AI infrastructure requires long-term stability. To support mission-critical deployments, the maintainers have introduced a strict formal deprecation policy:
- 12-Month Notice Period: At least 12 months must pass between when a feature’s formal deprecation is enacted and when the feature may actually be removed.
- Security Exception: A narrow exception exists solely for critical security vulnerabilities that require immediate mitigation.
This lifecycle commitment ensures enterprise architecture teams can upgrade MCP infrastructure on predictable schedules without fearing breaking protocol shifts.