Claude Code (gateway)
Claude Code (gateway)
Route Claude Code’s LLM calls through the Respan gateway for automatic logging, caching, and model switching. Only your RESPAN_API_KEY is needed.
Setup
Set environment variables
Add to your shell profile (.bashrc, .zshrc, or PowerShell $PROFILE):
Use ANTHROPIC_API_KEY, not ANTHROPIC_AUTH_TOKEN, for Claude Code through the Respan Anthropic gateway. Claude prefers ANTHROPIC_AUTH_TOKEN when both are set.
Or use Claude Code settings (persistent)
Add to ~/.claude/settings.json for global config, or .claude/settings.local.json in your project for repo-local config:
Setting ANTHROPIC_AUTH_TOKEN to an empty string clears any inherited token so Claude Code prefers ANTHROPIC_API_KEY. Run Claude Code as usual — all LLM calls will route through Respan.
The Anthropic gateway path is /api/anthropic/ (with trailing slash). For interactive Claude Code sessions, prefer shell env vars or ~/.claude/settings.json. If ANTHROPIC_AUTH_TOKEN is exported by your shell or terminal app, Claude may still prefer it, so unset or remove it first. On first interactive launch, Claude may prompt: Detected a custom API key in your environment. Choose Yes to use the Respan API key. If you skipped it earlier, run /config, search for custom, and enable Use custom API key. Non-interactive claude -p and claude --print use ANTHROPIC_API_KEY automatically.
Switch models
Once routed through the gateway, you can use any Claude model.
See the full model list.
Pin requests to a specific provider
To route Claude Code to your Vertex AI throughput (or any other provider) without changing the model slug, set the X-Respan-Route-Provider header. Claude Code keeps using its standard slug and the gateway resolves the provider transparently:
For a fallback across providers (e.g. primary Vertex, fall back to Anthropic on failure), use fallback_models instead. See Reliability for fallback_models and load_balance_group, and Providers & models for inline router controls (models, exclude_models, exclude_providers).