Claude Code Relay Setup for Live Token Streams

§1 Claude Code relay routing in plain English

Claude Code normally connects to its configured service to send requests and receive answers. An API relay sits between the coding tool and the model provider. Instead of changing your project code, you tell Claude Code where to send requests and which credential to present.

Environment variables are named settings that a running program can read. For this setup, one setting supplies the relay address and another supplies your relay credential. Claude Code inherits those settings when you launch it from the configured terminal.

Fluxio is a useful option if you want a streaming-first, pay-per-token relay. It supports Claude Code alongside other development tools, and its streaming responses arrive token by token rather than waiting for the entire answer to finish.

§2 The two environment variables that control relay access

• – ANTHROPIC_BASE_URL tells Claude Code which API server to contact instead of its default destination.
• – ANTHROPIC_AUTH_TOKEN supplies a credential as a bearer token when the relay uses that authentication method.

These settings change routing and authentication; they do not turn Claude Code into a general-purpose client for every model a relay offers. The relay must support the request format and features Claude Code expects, including streaming and tool use.

Fluxio’s API base is https://api.fluxioapi.com. OpenAI-style clients add /v1, but that instruction should not be copied automatically into a Claude Code configuration. Different client protocols can use different paths.

§3 Set Fluxio variables before launching Claude Code

In a Bash or Zsh terminal on macOS or Linux, the following example sets the relay address and a placeholder credential. Replace the placeholder locally with your Fluxio key, using the authentication method specified by Fluxio’s current Claude Code instructions.

$ export ANTHROPIC_BASE_URL="https://api.fluxioapi.com"

$ export ANTHROPIC_AUTH_TOKEN="YOUR_FLUXIO_KEY"

Then start Claude Code from that same terminal:

$ claude

The order matters. An already-running Claude Code session generally will not pick up environment changes made afterward. Exit it and launch it again after updating the variables.

These export commands affect the current terminal and programs started from it. They do not automatically configure every terminal, editor, or background process on your computer. Windows PowerShell uses different environment-variable syntax, so these commands are specifically for Bash and Zsh.

Keep your real key out of screenshots, repositories, and shared terminal logs. Typing a credential directly into an export command can also leave it in shell history. For regular use, prefer a trusted secret manager or a protected local configuration mechanism.

§4 Live token streams need a compatible connection

Streaming means the server sends pieces of a response while generation is still underway. Server-sent events, often abbreviated as SSE, are a common transport for these incremental updates. Claude Code handles the supported stream itself; you do not need to write an SSE reader.

A relay can preserve that responsive experience, but streaming is not a promise of instant output. Model processing, network latency, and intermediate proxies all affect when text appears. Some tool activity may also appear as status updates rather than continuous prose.

§5 Check relay authentication and routing errors

• – Authentication errors: Verify that the credential belongs to the relay and that its expected authentication method matches your configuration. Check for conflicting existing credentials, including ANTHROPIC_API_KEY, without printing their values.
• – Missing endpoint errors: Confirm the base URL against the relay’s Claude Code instructions. Do not append /v1 merely because an OpenAI SDK example includes it.
• – Unexpected destinations: Restart Claude Code from the terminal where you set the variables, and check whether another configuration source is overriding them.
• – Delayed or interrupted replies: Check connectivity, relay status, and any corporate proxy that might buffer or interrupt streaming connections.

§6 Fluxio credentials and coding data deserve care

Before using a relay with a private repository, review its data-handling terms and your organization’s policies. Coding requests can contain source code, file excerpts, and tool results. Use a dedicated credential where available, monitor token spending, and rotate any exposed key.

Sign up for Fluxio to try streaming replies in Claude Code.