Bug Description

Hi all,

Thanks for reading and for supporting n8n.

We've setup a python MCP server that can accept both SSE and HTTPStreamable connections;
I can successfully use it via n8n thanks to both the 'tool MCP' and 'Client MCP' nodes.

However, we have noticed a weird bug:

1. just run simple "List Tools" and "Execute Tool" requests via the mcp client node
2. wait for say 1 hour
3. upon restart of the server container, we see requests to the /sse endpoint in bulk (seem like one per each of the requests made at step 1.)

Looks like as if the n8n mcp client isn't shut down properly, but keeps trying reconnecting forever.
We don't know if we are missing some 'closing' message from our server side, but I would doubt about that since
with other tools (e.g. npx mcp-inspector) we do not see such behaviour.

To Reproduce

see the ticket

Expected behavior

no requests to /sse enpoint in bulk upon restarting the mcp server

Debug Info

Setup:
n8n under docker running Community Edition version and the mcp client nodes are:
"type": "n8n-nodes-mcp.mcpClient"
"typeVersion": 1

we are simply running the official docker image

docker.n8n.io/n8nio/n8n@sha256:f8fd85104e5a0cccb0d745536eeed64739c28239e6bfc2d0479380f31c394506
which was the 'latest' about 2 months ago

Operating System

Debian

n8n Version

1.92.2

Node.js Version

v20.19.0

Database

SQLite (default)

Execution mode

main (default)

Hosting

self hosted