Skip to main content
Claude Code is Anthropic’s terminal coding agent. It can be pointed at a custom endpoint, and people do point it at gateways, so it is a reasonable thing to try here. It will not work here yet, and this page says exactly why and exactly what it would take.

​
The one thing that decides it

Claude Code does not speak OpenAI chat completions. Anthropic’s own gateway documentation lists exactly three client formats it will use, and the one selected by a custom base URL is the Anthropic Messages API, posted to /v1/messages on whatever origin you give it. There is no OpenAI-compatible mode, and Anthropic states that it does not support routing Claude Code to non-Claude models through a gateway. So this is not a configuration problem. Pointing Claude Code at the base URL that works for every other harness on this site sends it to an endpoint this site does not publish, and nothing on these pages tells you what comes back. What would change it: a Messages endpoint on the published surface, in the tracker, shipped. When that lands this page stops being a warning and becomes a configuration page, and the settings below are the ones it will use.

​
What Claude Code would need, and where this API already agrees

Worth writing down, because two of these are already true here and are the parts people usually get wrong. The base URL is an origin, with no /v1. Claude Code appends /v1/messages itself, along with a query string. Every other harness on this site wants the /v1 included, so this one is the exception rather than the rule, and pasting the value from another page would be wrong twice over. The credential goes in ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY. The two are not aliases: ANTHROPIC_AUTH_TOKEN sends Authorization: Bearer, and ANTHROPIC_API_KEY sends x-api-key. This API reads Authorization: Bearer and nothing else, so ANTHROPIC_AUTH_TOKEN is the one that would be read here, and it is also the one Anthropic recommends for a third-party gateway. Authentication covers the header. Model ids would not be a problem. Anthropic states that Claude Code’s model id check runs only against Anthropic’s own API, and that behind a custom base URL the gateway defines the names, so any string is passed through. A slashed, provider-prefixed id is explicitly supported, which is the shape this API needs. There are several ids to pin rather than one: a session model, and separate family settings of which the small fast one drives background work. Token counting would not be a problem either. Claude Code will use a token-counting endpoint when one exists and count through the inference endpoint when one does not. Its absence costs requests, not correctness. Streaming would be mandatory, with keep-alive frames. Claude Code reads the stream as it arrives and gives up on one that goes quiet, counting keep-alive frames as traffic. A response assembled and sent in one piece stalls it.

​
What we did not verify

  • Anything about a Messages endpoint on this platform. This site publishes none, so this page makes no claim about one, including no claim about what a request to that path does today.
  • That any Claude Code session against this API succeeds or fails in any particular way. Nothing was run.
  • Whether Claude Code’s adaptive thinking field would be accepted. Claude Code sends it without a matching beta header for model names it does not recognise, which includes any id a gateway defines, and an endpoint that rejects it answers 400 naming the field. Claude Code has a setting that turns it off. We have not tested the interaction.
  • Which of the model settings a working configuration would actually need to pin. Anthropic’s documentation does not state a number and we are not going to invent one.