Error “Could not connect to your MCP server” when integrating external tool via SSE in AI Agent

Questions
1

I’m building an automation using n8n and trying to connect an AI Agent to an SSE (Server-Sent Events) endpoint using the MCP Client node.

However, I’m getting this error in the flow:

:cross_mark: Error in sub-node ‘MCP Client’

Could not connect to your MCP server

The endpoint I’m using is public and should work, but I keep getting the message:

“Cannot GET /mcp/teste/sse”

I’m not sure if the problem is with my endpoint configuration or something I’m missing on the n8n side.

Has anyone faced this before?

Any tips on what I should check to make this connection work properly with SSE?

Thanks in advance!

Captura de Tela 2025-04-12 às 16.10.07
Captura de Tela 2025-04-12 às 16.10.071986×610 82.2 KB

2 Likes
2

same issue

3

I got the same issue here. My MCP server was created in another workflow and my n8n is installed via coolify

1 Like
4

same issue. both production and test url did not work.

5

getting the same error got it to work for a while then it just would not connect ! anyone figured it out

6

Hey guys, this is a known issue.
It turns out MCP uses SSE connection and that’s a new thing for n8n.

So you have to adapt the infrastructure to treat the connection differently when using MCP.

With Coolify I found out you have to disable gzip compression for the /mcp path, while leaving it enabled for everything else.

That’s also true for traefik and nginx users.

Here’s some useful resources:

The video is in Portuguese, but it’s very well explained. Enable subtitles.

@Robin_Ramkhalawan @Lucas_Mozzatto_Chave @damiao_j @Jonathan_Politzki @khai-mle

5 Likes
7

Do you know where gzip is disabled for MCP? I can’t find that information easily.

1 Like
8

On the links I sent above I know there’s a guy teaching how to do it.

10

Hello everyone, I managed to connect my client by passing the following URL in the agent tool (SSE Endpoint field): http://[SERVICENAME]:[PORT]/mcp/… Mine ended up something like: http://n8n:5678/mcp/…
this is for using within n8n only but is better than connect via internet.

For those who wants to disable gzip just uncheck this and restart your n8n service:

image
image896×348 11 KB

1 Like
11

Did you fixed the issue…?? I am also having the same problem

12

same for this worfklow please help

1 Like
13

Thanks @damiao_j, using container name worked for me.

SSE Endpoint:

http://n8n:5678/mcp/{path}/sse

I also tested exposing n8n with ngrok and works fine too.

1 Like
14

I am having the same problem too, if someone has a concise and an easy-forward way to solve this issue please share it with the community :slight_smile:

2 Likes
15

Hey @yasafa, please provide more details and what you have tried so far. Is it local or hosted?

In the meantime, check for any typos in scheme, URL, port, and path.
{SCHEME}://{URL}:{PORT}/{PATH}

16

no i mean it was really similar to what others described, when connecting an MCP client with a local MCP trigger (on n8n) it fails to connect using the production URL. is there another way to do it?
because with external mcp servers, like brave search I managed to connect with STDIO, so yeah I was wondering if there is a simple solution to connect the client to a local MCP trigger on n8n itself

1 Like
17

I simply replaced “localhost” with the IP equivalent 127.0.0.1 and it worked

1 Like
18

Where did you do this?

1 Like
19

You enter that on MCP Clients’s SSE Endpoint field.

2 Likes
20

SOLVED: In my case (n8n is hosted in hostinger) the solution was:

1- add the port after my domain
2- due n8n doesn’t have SSL, the url of my MCP Server is http://
3- the example: http://n8n.YOURDOMAIN:5678/mcp/XXXXXXXX/sse

Hoping it helps you.

21

Thank you. Worked for me as well.

1 Like