Installing Grafana MCP Helm chart on EKS with Ingress failing connection

Monitoring and Observability Grafana
1

Hey Folks! Just installed Grafana’s MCP helm in the EKS cluster and hosted it with ingress.
On requesting from browser this is what I see:

In my N8N MCP Client tool I see following error
I know that connectivity is there. But IDK why I cannot connect to it.
Can anyone please help?

2

can you please share the details of your setup on the n8n side?

3

specifically the mcp client setup?

4

Sure. Here you go :pray:
AFAIK All I need to do is provide the service endpoint with port.
I also tried it with Ingress.
I didn’t provide anything for Auth because I already provided Grafana URL and Service Account Token in the helm env values.

What I did notice that it works inconsistently. It worked only once with the SSE Endpoint where it was able to fetch all the tools and I was able to choose the Tools in the “Tools to include” but then it stopped working again when the workflow was actually called.

I would be very happy to know that my configuration was incorrect :sweat_smile:

5

Also this is what I get from the SSE endpoint, which is something to be expected?

6

Maybe might have some hints as he has done some nice videos about mcp

7

What worked for me using vscode mcp client and cursor ide mcp client was pointing them to my grafana url and not to mcp endpoint

From my understanding grafana is kind of like a broker, or middle man

Mcp uses a configuration that points to grafana url and token

And the client also uses same

8

Is there a way to configure the same in n8n?

9

Well, point it to grafana url and provide it the service account token. That is what I was pointing out in previous post

10

But that means I am not using Mcp but the Grafana application with the agent no?

11

Ahh you mean to say that the token which I got in the browser I should add that as an authorisation token in the Mcp client?

12

https://github.com/grafana/mcp-grafana?tab=readme-ov-file#usage|https://github.com/grafana/mcp-grafana?tab=readme-ov-file#usage
This should help answer that

13

what kind of node did you create to use the mcp server?

14

I used the MCP client

15

it helps if you provide the details on that plugin.

https://www.npmjs.com/search?q=keywords%3An8n-community-node-package

16

Maybe I’m running an older version of n8n which doesn’t support this anymore
Let me try a different version

17

best approach I would recommend is do it all in docker first.

18

I guess the helm is not usable. I’ll have to try with an MCP node that helps me run a command and then try to connect to the grafana URL just like you did in the Cursor.
But my only question is why?
Technically it should work.
I tried the same in Docker compose and it still doesn’t work with the latest version
Both on the SSE and HTTP streamable

19

If you want to re-produce it, run n8n on your local:

docker run -d -p 5678:5678 n8nio/n8n

Then run the Docker container for your MCP server
In N8N search for the Agent node:

In the agent node you can try adding the MCP tool:

And then in the MCP client tool you can add the Grafana mcp server endpoint URL

20

tried that hit a wall. there seems to be an issue with docker mcp grafana

with binary mcp grafana running stdio
vscode works
cursor works

just by providing these config file

{
“servers”: {
“mcp-grafana”: {
“type”: “stdio”,
“command”: “mcp-grafana.exe”,
“args”: [
“-t”,
“stdio”
],
“env”: {
“GRAFANA_URL”: “http://localhost:1200”,
“GRAFANA_SERVICE_ACCOUNT_TOKEN”: “zoozoo”
}
}

},
"inputs": []

}