What is the main scaling tradeoff introduced by stateful MCP sessions?

Attempt it

Quick reference

TL;DR

A stateful MCP session pins a client to one server instance, so the load balancer needs sticky routing, which blocks the stateless horizontal scaling that cloud replicas rely on.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Think of a coat check at a theater. The first time you arrive, you hand over your coat and get a numbered ticket; the attendant remembers which hook holds your coat for the whole evening. That memory is convenient: you do not re-describe your coat every time. But it means you must return to that exact attendant. If the theater hires ten attendants and sends you to a random one, only the one holding your ticket can find your coat. A stateful MCP session works the same way. The server remembers your session in its own memory after one setup handshake, so later messages get there fast. The cost is that every later message has to come back to that same server; you cannot just spray requests across a fresh pool of identical servers.

Key concepts

Stateful sessions Sticky routing Horizontal scaling Capability negotiation

Quick scan · 2–4 min

Everything important, quickly.

4 min: session lifecycle and capability negotiation, what the persistent channel buys, why in-memory state forces sticky routing, the conflict with stateless scaling, and the stateless-mode or shared-store mitigations.

Side-by-side

Concern	Stateful MCP session	Stateless request-response
Capability negotiation	Once at initialize, reused	Implied or repeated per call
Notifications, subscriptions	Yes, pushed over live channel	No, client must poll
Routing requirement	Sticky, affinity to one instance	Any replica, fully interchangeable
Horizontal scaling	Constrained, pinned traffic	Native, spread freely
Failure on instance loss	Session dies, must reconnect	Retry hits any healthy replica

Real-world examples

Real products, models, and research that use this idea.

Claude Code keeps a live MCP session per server so tools/list_changed and resource subscriptions push updates without re-handshaking.
The Streamable HTTP transport, the 2025 successor to HTTP+SSE, adds a stateless mode precisely so remote MCP servers can scale across replicas.
Teams hosting remote MCP servers behind AWS ALB or NGINX enable session affinity, or externalize session state to Redis, to survive autoscaling and rolling deploys.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow would you make a remote MCP server horizontally scalable while still supporting subscriptions?

Sign in or go Pro to see the approach for this follow-up.

QWhat does the Streamable HTTP transport change versus the old HTTP+SSE transport for scaling?

Sign in or go Pro to see the approach for this follow-up.

QHow do you handle a client that disconnects without cleanly closing its session?

Sign in or go Pro to see the approach for this follow-up.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.