How do I use Portkey with Claude Code?
Put ANTHROPIC_BASE_URL=https://api.portkey.ai in ~/.claude/settings.json, add your Portkey key and a provider header such as @anthropic-prod in ANTHROPIC_CUSTOM_HEADERS, and set the model names explicitly. Tokens bill to the provider behind that integration.
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.portkey.ai",
"ANTHROPIC_AUTH_TOKEN": "YOUR_PORTKEY_API_KEY",
"ANTHROPIC_CUSTOM_HEADERS": "x-portkey-api-key: YOUR_PORTKEY_API_KEY\nx-portkey-provider: @anthropic-prod",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-20250514",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4-20250514",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-20250514"
},
"model": "claude-sonnet-4-20250514"
}
Portkey's examples still name Claude 4 models; replace them with current IDs from your Model Catalog. ~/.claude/settings.json keeps the values for every session. Run /status in Claude Code to check the base URL.
Setup in four steps
- In Portkey's Model Catalog, add a provider integration with your Anthropic, Bedrock or Vertex credentials. Its slug, such as @anthropic-prod, is what Claude Code will name.
- Create a Portkey API key.
- Paste one of the blocks above into ~/.claude/settings.json, with no /v1 on the base URL.
- Start Claude Code and send a prompt; the request should appear in Portkey's logs. A lightning symbol there marks a passthrough request.
Gotchas in Portkey's own docs
| Issue | What happens | What to do |
|---|---|---|
| /v1 in the base URL | "API Error: 500 fetch failed" | Use https://api.portkey.ai exactly |
| No model names | Requests fail | Set ANTHROPIC_DEFAULT_SONNET_MODEL, _OPUS_ and _HAIKU_ to IDs your provider serves |
| Old example IDs | The docs' examples name Claude 4 models from May 2025 | Replace them with current IDs from your Model Catalog |
| Beta features | Anthropic beta headers are dropped unless forwarded | Add forward_headers: ["anthropic-beta"] to your Portkey config |
| Other models | Swap the provider slug, for example @openai-prod, and set ANTHROPIC_MODEL | Test tool use on a real task; some models handle it worse than Claude |
| Claude Max sign-in | Documented under Enterprise; costs show as $0 in Portkey | Use it for logs and limits, not billing |
From Portkey's Claude Code docs, October 13, 2026. Portkey also recommends running the latest Claude Code release.
What it costs
Portkey bills logs, not tokens: free for 10,000 logged requests a month, then $49 on Production. The tokens go to your Anthropic, Bedrock or other provider account at its own price, which for a typical coding-agent day is $3.53 on Claude Sonnet 5.5 and $7.06 on Opus 5.5 at list price, measured in our Agent Burn Index. Portkey can cap a key's spend on Enterprise and for select Pro customers; it does not pick a cheaper model for routine steps.
The catch
Portkey's setup has four moving parts: a provider integration, a key, a custom header and three model variables. Each is a place where a typo turns into a 500, so change one at a time and watch the logs.
Where Standard Compute fits
Standard Compute needs two of those values, the base URL and a key, and keeps Claude Code's own model names. It sits in front of Claude Code like Portkey does, and also works on the bill: routine steps go to efficient models, inside a plan from $19 that stops at its budget. If your company already governs every LLM call through Portkey, stay there.
FAQ
What is the Portkey base URL for Claude Code?
https://api.portkey.ai, with no /v1, because Claude Code adds /v1/messages itself. Portkey's troubleshooting table lists the /v1 suffix as the cause of "500 fetch failed".
Can Claude Code use GPT or other models through Portkey?
Yes. Portkey documents swapping the provider slug in ANTHROPIC_CUSTOM_HEADERS, for example to @openai-prod, and setting ANTHROPIC_MODEL; Portkey translates the request. How well a model handles Claude Code's tools varies.
Does Portkey work with Claude Code on Bedrock or Vertex?
Yes. Point the provider header at a Bedrock or Vertex integration (Portkey's examples use @bedrock-prod and @vertex-prod) and use that platform's model IDs.
Does Portkey work with Codex CLI?
Yes. Portkey documents a model provider with base_url https://api.portkey.ai/v1 and env_key PORTKEY_API_KEY, with wire_api set to chat by default or responses, and models named @provider-slug/model.