[Feature Request] Support for Claude options

[MKr4zy]

I host my own llm models, and give them to Claude via a seperate Settings.json file. I would be nice, If there would be support for Options like this.

[marcus]

Hey @MKr4zy! Starling here (AI assistant on the project). 👋

Thanks for the request! A bit more context would help us understand what you're after:

• Are you referring to Claude's MCP server config (via claude_desktop_config.json or a similar Settings.json), or are you passing custom options to the Claude CLI itself (e.g., --model, API endpoint, custom headers)?
• Which Claude integration are you using? (Claude CLI, Claude Code, opencode, or something else?)
• What specific options are you configuring in your Settings.json that you'd want sidecar to pass through or respect?

Surfacing this to @marcus once we have a clearer picture of the use case. Thanks for taking the time to report! ✦

[MKr4zy]

I'm configuring a seperate .json file in my claude folder, where I set the IP of my lokal LLM Server. I'm using LM Studio to host them, but that shouldn't make a difference. I'm calling claude code like this: claude --settings <path_to_config.json>.

Example Config:

"env": {
    "ANTHROPIC_BASE_URL": "http://xxx.xxx.xxx.xxx:1234/",
    "ANTHROPIC_AUTH_TOKEN": "dummy",
    "API_TIMEOUT_MS": "3000000",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1,
    "ANTHROPIC_MODEL": "default_model",
    "ANTHROPIC_SMALL_FAST_MODEL": "default_model",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "default_model",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "default_model",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "default_model"
}

[marcus]

Hey @MKr4zy! Thanks for the detailed follow-up — that clears things up a lot.

So the use case is: you're running a local LLM server (LM Studio) and pointing Claude Code at it via a custom --settings JSON that overrides the API base URL, auth token, and model names. You want sidecar to respect or pass through that config when it launches Claude Code.

That's a totally reasonable ask. The specific question for @marcus is whether sidecar should:

1. Support passing a --settings <path> flag to Claude Code when launching agents (so your custom LM Studio config gets loaded automatically), or
2. Pass through a set of env vars (ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, etc.) to the Claude Code process sidecar spawns

Flagging this for @marcus to decide on the right approach — it has implications for how sidecar manages agent launch configs. In the meantime, one potential workaround: if you launch sidecar itself with those env vars already set in your shell, Claude Code should inherit them. Does that work for your setup? ✦

[MKr4zy]

As I currently have a License for Claude Code, I want to be able to use both of these options. So in my opinion a better approach would be a way to configure custom agents in sidecar.

[marcus]

Thanks for the direction, @MKr4zy — that reframes the request nicely. Rather than just a flag passthrough, you're describing a first-class concept: custom agent definitions in sidecar, where each agent entry can carry its own launch args and config (like a --settings path pointing to your LM Studio config).

That's a more composable approach than either option I mentioned. Surfacing this framing to @marcus — custom agent support would also address some of the adjacent requests around supporting non-Claude agents. ✦