Hosting the mcp server locally on another server in the network

[maly-miller]

Hi all,

I am hosting the mcp server in my local network in a different sever than my n8n instance and also different from my workstation that has Claude Desktop. Can someone help me with a the json to connect. I am a bit confused as all what is provided in the repo is doesnt really provide any context about using a url and a bearer token and how to generate them.
My n8n-mco server is running on server with IP 192.168.1.185. tried to use https://1923.168.1.185/mcp but i cant even reach it

[czlonkowski]

Hey @maly-miller — few things going on. Lets unstick you.

First, the URL. You wrote https://1923.168.1.185/mcp — two problems:

1. Typo (1923 instead of 192).
2. Missing port. The HTTP server listens on port 3000 by default, not 80/443. So the URL should be http://192.168.1.185:3000/mcp. Plain http, not https — there is no built-in TLS (you would need nginx/Caddy in front if you want HTTPS).

Second, the token. You generate it yourself, it is not given to you. Something like:

openssl rand -base64 32

Save the output, pass it to the server as AUTH_TOKEN. Minimum 32 characters or it will refuse to start.

Third, starting the server. On 192.168.1.185:

AUTH_TOKEN=your-long-token-here npm run start:http

It binds to 0.0.0.0:3000 by default, so it is reachable from the LAN immediately. Hit http://192.168.1.185:3000/health from your workstations browser — if you see a response, the server side is done.

Fourth, Claude Desktop. This is the part the docs could explain better: Claude Desktop only speaks stdio natively. For an HTTP MCP on another machine, you use mcp-remote as a bridge. Put this in your Claude Desktop config:

{
  "mcpServers": {
    "n8n-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://192.168.1.185:3000/mcp",
        "--header",
        "Authorization: Bearer your-long-token-here"
      ]
    }
  }
}

Restart Claude Desktop and the tools should appear. Full HTTP deployment guide lives in docs/HTTP_DEPLOYMENT.md if you want the long version.

Ping me if something still does not connect — paste the output of curl -i http://192.168.1.185:3000/health from your workstation and we will narrow it down.