Microsoft has released version 2.0 of its official C# SDK for Model Context Protocol servers and clients, updating it to match the protocol’s latest revision and its move to stateless communication by default. The new release also adds support for HTTP standardization and Multi Round-Trip Requests, making MCP easier to deploy through normal web infrastructure.
What Changed
The biggest shift in the 2026-07-28 MCP revision is that stateless operation is now the default. In the C# SDK, that means HTTP server transport now starts in stateless mode rather than stateful mode, so developers no longer need to build around long-lived sessions for basic MCP traffic.
Microsoft says the revised HTTP surface also makes MCP requests more self-describing. A call such as tools/call now carries headers like Mcp-Method and Mcp-Name, which allows load balancers, proxies, gateways, and web application firewalls to understand and route traffic without inspecting the request body.
Why It Matters
This is a practical change for teams building MCP services in real environments. Stateless requests are easier to scale, simpler to route, and generally fit better into existing enterprise infrastructure.
The update also lowers the friction for security and operations teams, since normal HTTP tooling can now handle more of the MCP traffic pattern directly.
MRTR Support
Version 2.0 also adds support for Multi Round-Trip Requests on both the server and client sides. MRTR is meant for cases where a tool cannot finish in a single exchange, such as when a server needs user confirmation before performing an action.
Previously, those interactions depended on an active session. With MRTR, the continuity is carried in the payload instead, so the exchange can happen without a persistent connection.
What’s Next
Microsoft says the next major focus for the SDK is end-to-end authentication and authorization. The company wants secure MCP deployments to be as easy to set up as the rest of the stack, building on the spec’s closer alignment with OAuth and OpenID Connect.
That suggests MCP is moving from an experimental integration layer toward something more enterprise-ready, especially for AI applications that need secure access to external tools and services
