Multi round-trip requests (MRTR)
What it is
Multi Round-Trip Requests (MRTR) is an MCP protocol pattern that lets a server ask a client for more information mid-operation without holding a connection open. Instead of blocking on a live bidirectional stream, the server replies to an incomplete request with an "input required" result carrying serialized state; the client gathers what's needed and retries the same request with that extra information attached, repeating until the server has enough to finish.
Why it's emerging now
It's part of MCP's shift, via SEP-2322 in the 2026-07-28 spec release candidate, from a stateful, bidirectional-stream protocol toward a stateless request/response one. Features like sampling and elicitation used to require servers to keep a connection — and often a session store like Redis — open while waiting on the client; MRTR lets those same interactions happen over plain stateless HTTP, which matters for MCP servers that need to run behind ordinary load balancers or scale horizontally in the cloud. Production servers such as the GitHub MCP Server have already adopted it and dropped their session-storage layer as a result.
How it relates to what you know
It's a mechanism inside MCP (already known), specifically for the server-initiated request patterns — elicitation and sampling — that previously needed a persistent connection; it's the protocol-level fix that makes those patterns compatible with stateless deployment.