Docs

Two settings. Any agent.

Every tool needs the same two things: a CodingPass Base URL and your key. Paste the key once and every snippet on this page is ready to copy.

Base URLs

https://codingpass.ai

Anthropic-compatible

Messages API. Claude Code, Cline and the Anthropic SDKs. No /v1 at the end.

https://codingpass.ai/v1

OpenAI-compatible

Chat Completions and Responses. Codex, OpenCode and the OpenAI SDKs.

Your setup

Paste your key, pick your tool.

API keyGet a key

Stays in this tab. It only fills in the snippets on this page.

Model

Each tool starts with a sensible default. Your plan decides which models answer.

Add the two export lines to ~/.zshrc or ~/.bashrc, open a new shell, then start Claude Code.

~/.zshrc

export ANTHROPIC_BASE_URL="https://codingpass.ai"
export ANTHROPIC_AUTH_TOKEN="cp-live-your-key"

Then run

claude --model claude-sonnet-5-5

Full Claude Code guide

Check that the key works

List the models your key can reach. A JSON list back means the key and the URL are right; a 401 means the key is wrong.

Terminal

curl https://codingpass.ai/v1/models \
  -H "Authorization: Bearer cp-live-your-key"

Your tool isn’t listed?

Any tool that lets you set an Anthropic or OpenAI Base URL works. Look for a Base URL or custom provider field, paste the URL and your key, and use a model ID from Models.

Still stuck? Write to [email protected] with the tool name and the x-request-id header from the response.

Errors

Every error, its status, and what to do.

The gateway answers with a standard HTTP status and the error shape your client already knows. The type is the one Anthropic and OpenAI use, so SDKs handle it as usual; the message is written for people. The error code below shows next to each failed request in the console’s request log.

Two shapes, depending on the endpoint

Anthropic-compatible endpoints (/v1/messages, /v1/messages/count_tokens)

HTTP/1.1 429 Too Many Requests
retry-after: 4253
x-should-retry: false
x-ratelimit-remaining-5h: 0
x-ratelimit-reset-5h: 1h10m53s
x-request-id: req_8f2c41d07ab3

{
  "type": "error",
  "error": {
    "type": "rate_limit_error",
    "message": "Account limit reached: your 5h quota is used up for all models. Resets at …"
  }
}

OpenAI-compatible endpoints (/v1/chat/completions, /v1/responses, /v1/models)

HTTP/1.1 401 Unauthorized
x-request-id: req_5b1f09c2e7d4

{
  "error": {
    "message": "Invalid API key",
    "type": "authentication_error",
    "param": null,
    "code": null
  }
}

Gateway errors by HTTP status
Status	Code	When it happens and what to do
400Bad request
400	invalid_requestinvalid_request_error	Request body not valid When it happens: The body is not valid JSON, or a required field is missing: `model`, `messages`, or `max_tokens` on the Anthropic API. How to fix: Update your client. Old versions sometimes send fields in an outdated shape. If you call the API from your own code, compare the body with the provider’s API reference.
401Authentication
401	invalid_api_keyauthentication_error	API key not accepted When it happens: The key is missing, mistyped, deleted, or sent in a header the gateway does not read. How to fix: Copy the key again from Keys in the console, with no spaces or line breaks. Anthropic clients send it as `x-api-key`, OpenAI-compatible clients as `Authorization: Bearer`. After deleting or rotating a key, update every client that still uses the old one.
403Permission
403	no_subscriptionpermission_error	No active plan When it happens: The key is valid, but its account has no active plan: none was bought, it has ended, or a renewal payment failed. How to fix: Check the plan and payment state in Billing. No plan yet? Pick one on Pricing. The same key works once it is active.
404Not found
404 / 405	wrong_base_urlnot_found_error	Wrong Base URL or method When it happens: The request reached the gateway on a path or method it does not serve. Usually the Base URL has an extra or missing `/v1`, or a browser opened the API address as a web page. How to fix: Anthropic clients such as Claude Code take the Base URL without `/v1`; OpenAI-compatible clients add `/v1`. The error message shows both values. Copy the Base URL from your tool’s setup guide instead of typing it. Send `POST` requests. The API has no web page to open in a browser.
404	model_not_foundinvalid_request_error	Model not in your plan When it happens: The model ID is misspelled, or the model is not included in your plan. How to fix: Copy the exact model ID from Models. Check that your plan includes the model. Higher plans add models.
413Too large
413	body_too_largeinvalid_request_error	Request too large When it happens: The request body is over the gateway’s size limit, usually because of very long context or large attachments. How to fix: Compact or clear the conversation (`/compact` in Claude Code) and retry. Attach fewer or smaller files and images.
429Limits
429	quota_exceededrate_limit_error	Allowance used up When it happens: A 5-hour, weekly or monthly cap is reached, for one model or for the whole account. The message names the window and the reset time; `retry-after` gives the seconds to wait. How to fix: Wait until the reset time in the message. If only one model hit its cap, switch to another model. The others keep working. Use an allowance reset in Billing. A full reset clears 5-hour, weekly and monthly usage; a weekly reset clears 5-hour and weekly usage. Upgrade on Pricing for higher caps. Usage limits explains how the windows work.
429	concurrency_limitedrate_limit_errorSafe to retry	Too many requests at once When it happens: Your plan allows a fixed number of requests in flight, and a new one arrived while all of them were busy. How to fix: Run fewer agents or subagents in parallel. Retry after a second. The response carries `retry-after: 1`. Higher plans allow more concurrent requests. Compare them on Pricing.
429	no_upstream_availablerate_limit_errorSafe to retry	Model at capacity When it happens: This model is at capacity right now. How to fix: Wait the number of seconds in the `retry-after` header, then retry. Switch to another model in your plan for now.
5xxModel provider
4xx / 5xx	upstream_errorapi_errorSafe to retry	Model provider returned an error When it happens: The model provider rejected or failed the request, and the gateway passed its status and message through. A 502 means the request still failed after retries. How to fix: Read the provider’s message. A 4xx usually points at something in the request to change. Retry a 5xx after a few seconds. If it keeps failing, check Status.
503Unavailable
503	plan_unavailableapi_errorSafe to retry	Plan check unavailable When it happens: The gateway could not load your plan for a moment. The problem is on our side, not with your key. How to fix: Retry after a few seconds. If it lasts more than a few minutes, check Status and write to [email protected].
503	quota_unavailableapi_errorSafe to retry	Allowance check unavailable When it happens: The gateway could not read your usage counters, so it refused the request instead of guessing. The problem is on our side. How to fix: Retry after a few seconds. If it lasts more than a few minutes, check Status and write to [email protected].
503	upstream_unavailableapi_errorSafe to retry	Request could not be served When it happens: The model could not take this request before replying. Retrying usually works. How to fix: Retry after a few seconds. If it lasts more than a few minutes, check Status and write to [email protected].
503	no_upstreamapi_error	Model not served right now When it happens: The model is in your plan but is not available at the moment. How to fix: Switch to another model in your plan for now. If it lasts more than a few minutes, check Status and write to [email protected].
200 / 499During the reply
200	stream_incompleteapi_errorSafe to retry	Stream cut off When it happens: The reply started with 200 but the stream ended before it was complete, so the client shows a reply that stops partway or an error at the end. How to fix: Retry after a few seconds. Check proxies or VPNs between you and the API that cut long connections.
499	client_aborted	Cancelled by the client When it happens: Your client closed the connection before the reply finished: you pressed Esc or Ctrl+C, the client timed out, or the network dropped. 499 appears only in your request log; the client never receives it. How to fix: If you cancelled the request, there is nothing to fix. If it happens on its own, raise the client’s timeout. Long replies can take several minutes.

Errors marked “safe to retry” are temporary: wait a second or two, or the retry-after seconds, and send the same request again. Most SDKs do this on their own for 429 and 5xx. Everything else needs a change on your side first.

Useful response headers

Response headers the gateway adds
Header	Meaning
x-request-id	Unique ID of this request, the same one shown in your usage log. Include it when you write to [email protected].
retry-after	Seconds to wait before retrying. Sent with 429 and with temporary 503s.
x-should-retry	Sent as `false` with `quota_exceeded`: the official Anthropic and OpenAI SDKs then stop retrying on their own, because the wait is a whole window.
x-ratelimit-remaining-<window>	Sent as 0 with `quota_exceeded`. The suffix is the window you hit: `5h`, `week` or `month`. Whether it's the account or a single model is in the error message.
x-ratelimit-reset-<window>	Sent with `quota_exceeded`: how long until that window frees up, for example `1h10m53s`.
x-ratelimit-limit-concurrency	Sent with `concurrency_limited`: how many requests your plan runs at once. `x-ratelimit-remaining-concurrency` is 0. Not sent when the 429 is about upstream capacity.