Can your MCP server survive a load balancer?

Moved behind a load balancer, and a long tool call vanished mid-run.

Intermediate to advancedRead 10 minDoing: an afternoon8 steps

▶ Start

What you'll have at the end

The same MCP server keeping one session correct across three separate running instances, checked with a real load test

Before you start

You need

An MCP server already reachable over Streamable HTTP instead of only as a local subprocess, with at least one tool call that can legitimately run longer than an ordinary web request.

Not covered

Choosing or configuring a specific load balancer or container platform, and the fully stateless, signed-token tier for fleets that don't yet run at that scale.

Leans on

8 steps

Where to go next

Move a stdio MCP server onto HTTP

Get the server onto Streamable HTTP first; putting a load balancer in front only matters once there's a remote endpoint for it to sit behind.

Checked 15 Aug 2026

Part of the MCP Servers Implementation cookbook