Configuration Parameters for Remote MCP Servers?

[tkaria]

How are people configuring remote servers today? It's pretty straightforward with a local server - e.g:

When working with local MCP servers, configuration is straightforward - we can pass parameters via command-line arguments:

Local

{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["server.js", "--api-key=${API_KEY}", "--foo=bar"]
    }
  }
}

but it seems there's no standard for remote servers:

Remote

{
  "mcpServers": {
    "remote-server": {
      "url": "https://example.com/mcp"
      // 🤔 How do we pass configuration here?
    }
  }
}

I can imagine multiple approaches to addressing this including:

1. Query parameters
   e.g. `url: https://example.com/mcp?foo=bar"
2. Custom headers
   e.g. x-mcp-configuration: foo=bar;zip=flip
3. Post-connect negotiation
   e.g. Adding an optional step pre or post initialization to specify config params

What I'd like to understand:

1. How are people solving this today?
2. What configuration needs do you have?
3. Should this be standardized?
4. How would we handle secure config params (e.g. API keys, secrets)
5. How do clients discover configuration options?

[jonathanhefner]

Check out the server.json format in modelcontextprotocol/registry! 😃