Setup

Sign in and register the Nuon MCP server with your LLM client.

Prerequisites

Sign in. This writes the token every setup below reads:

nuon auth login    # writes api_token to ~/.nuon

nuon agents help prints this page, checked against your own config. When an agent runs that command, it prints markdown instead.

Connect

Recommended: register the stdio proxy so the client runs nuon agents mcp --allow-writes. Token and org stay in ~/.nuon, so nothing secret goes into the client config. The proxy derives its URL from the configured API hostname (api.<hostname> becomes mcp.<hostname>/mcp).

Setup (stdio)

Nuon MCP uses stdio. The client starts a local process and exchanges MCP messages on that process’s standard input and output. Every MCP connection should spawn nuon with arguments agents, mcp, and --allow-writes.

Claude Code

claude mcp add --transport stdio nuon -- nuon agents mcp --allow-writes

Or .mcp.json in a project:

{
  "mcpServers": {
    "nuon": {
      "command": "nuon",
      "args": ["agents", "mcp", "--allow-writes"]
    }
  }
}

Cursor

Cursor has no mcp add. Save this as ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), then enable it:

{
  "mcpServers": {
    "nuon": {
      "command": "nuon",
      "args": ["agents", "mcp", "--allow-writes"]
    }
  }
}

agent mcp enable nuon

Amp

amp mcp add nuon -- nuon agents mcp --allow-writes

# this workspace only
amp mcp add nuon --workspace -- nuon agents mcp --allow-writes

Or ~/.config/amp/settings.json (user) / .amp/settings.json (workspace). Amp’s key is amp.mcpServers:

{
  "amp.mcpServers": {
    "nuon": {
      "command": "nuon",
      "args": ["agents", "mcp", "--allow-writes"]
    }
  }
}

Other

Use that client’s MCP docs for the file path and JSON key. Point it at the nuon binary with arguments agents, mcp, and --allow-writes.

Overriding the MCP URL

The proxy derives its upstream URL from api_url in your CLI config when the hostname starts with api. or app. (https://api.nuon.co and https://app.nuon.co both resolve to https://mcp.nuon.co/mcp). Pass --url when the MCP URL does not follow from the API URL, for example a self-hosted or Nuon BYOC control plane, or any deployment where the two hostnames differ. To change the name in the client’s MCP list, pass --name. Put those flags on the command the client runs, next to --allow-writes:

nuon agents mcp --allow-writes --url https://mcp.example.nuon.co/mcp --name nuon-example
claude mcp add --transport stdio nuon -- nuon agents mcp --allow-writes --url https://mcp.example.nuon.co/mcp --name nuon-example
amp mcp add nuon -- nuon agents mcp --allow-writes --url https://mcp.example.nuon.co/mcp --name nuon-example

For a client configured by file, extra flags go inside args. Use that client’s JSON key (examples above):

{
  "mcpServers": {
    "nuon": {
      "command": "nuon",
      "args": ["agents", "mcp", "--allow-writes", "--url", "https://mcp.example.nuon.co/mcp"]
    }
  }
}

A non-default CLI config (nuon -C /path/to/config agents mcp --allow-writes) carries its own token, org, and api_url. Put -C and --allow-writes on the registered command (in args when the client is configured by file).

Run nuon agents help to confirm which MCP URL resolves before registering.

After the client is connected, paste a first prompt from Sample queries.