---
title: Use Anthropic SDKs with a Custom Base URL
description: "Set the Anthropic SDK base URL to router.apitoken.sale: same Python and TypeScript code, same model IDs, and a flat 50% off every token."
url: https://apitoken.sale/docs/learn/anthropic-sdk-base-url
language: en
---

# Point the Anthropic SDK at apiToken.sale

Every official Anthropic SDK accepts a custom Anthropic SDK base URL, so moving to apiToken.sale is a one-argument change. Your model IDs, message code and streaming logic stay exactly the same — only the endpoint and the per-token price change.

## One argument switches the endpoint

Both official Anthropic SDKs — Python and TypeScript — let you override the API root when you construct the client. Set it to https://router.apitoken.sale and every request your code already makes is served by apiToken.sale's gateway instead of api.anthropic.com. Nothing else in your codebase moves: same anthropic package, same Messages API, same model IDs like claude-opus-4-8, same response objects.

What changes is billing. Each call is metered at official Anthropic token rates, your flat 50% discount is subtracted, and the net amount is drawn from a prepaid balance you top up in whole-dollar amounts. No subscription, no per-seat fee — idle days cost nothing.

## Python: base_url on the client

```
from anthropic import Anthropic

client = Anthropic(
    base_url="https://router.apitoken.sale",
    api_key="sk-pool-•••",
)
msg = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
```

The async client takes the identical keyword: AsyncAnthropic(base_url=..., api_key=...). Streaming via client.messages.stream, tool use, system prompts and prompt caching all ride on the same connection — there is no separate endpoint to configure for them.

> Pass the bare root, not a path. The SDK appends /v1/messages itself, so base_url=".../v1" produces requests to /v1/v1/messages and a 404. The same rule applies to the TypeScript SDK.

## TypeScript: baseURL on the client

```
import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://router.apitoken.sale",
  apiKey: "sk-pool-•••",
});
const msg = await client.messages.create({
  model: "claude-opus-4-8",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});
```

The @anthropic-ai/sdk package sends the x-api-key and anthropic-version headers for you, exactly as it does against the official endpoint. Retries, timeouts and error classes (APIError, RateLimitError and friends) behave identically, so existing error handling keeps working.

## Prefer environment variables in shared code

Both SDKs read ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY from the environment when the constructor arguments are absent. That makes the switch a deployment detail instead of a code change — useful when the same repository runs against different endpoints in development and production.

```
export ANTHROPIC_BASE_URL=https://router.apitoken.sale
export ANTHROPIC_API_KEY=sk-pool-•••

# your code now constructs Anthropic() with no arguments
```

Tools built on top of the SDK inherit the same variables. Claude Code, for example, honours ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY directly, and frameworks like LangChain or LiteLLM forward the same environment to their Anthropic client underneath. Explicit constructor arguments win over environment variables when both are set, so a one-off override in a script never leaks into your deployed configuration.

## What crosses the gateway unchanged

- The full Messages API surface: POST /v1/messages with the same request and response JSON.
- SSE streaming — incremental chunks arrive exactly as from api.anthropic.com.
- Tool use and function calling, including multi-turn tool_result loops.
- System prompts, vision inputs and prompt caching with cache_control breakpoints.
- The usage object on every response, so your token- and cost-tracking code keeps working.
- Model IDs: claude-opus-4-8, claude-sonnet-5, claude-haiku-4-5 and the rest of the supported catalog.

One key covers every supported model — Claude alongside GPT, Gemini and Kimi — so a multi-provider project keeps a single credential and a single balance. Per-request spend and the applied discount are visible in the dashboard after each call.

[Supported model IDs and per-model pricing](/models)

[Estimate monthly spend in the cost calculator](/tools/claude-api-cost-calculator)

## First-request checklist and common errors

1. Create a free account, open the dashboard and generate a key — it looks like sk-pool-… and works across the supported Claude, GPT, Gemini and Kimi models.
2. Set base_url / baseURL to https://router.apitoken.sale in code, or export ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY.
3. Run the Python or TypeScript snippet above once and confirm you get a normal Anthropic message response.
4. Open the dashboard and verify the request appears with its token usage, cost and discount.

| Status | Meaning | Fix |
| --- | --- | --- |
| 401 Unauthorized | Missing or wrong x-api-key, or wrong base URL | Re-check the key and that the URL is the bare root |
| 400 Bad Request | Malformed body, or documented_limitation / unsupported_parameter on a named field | Check the model ID and max_tokens; omit the named field if details.error_code is set |
| 402 billing_error | Insufficient prepaid balance — error.type billing_error, not a 429 | Top up any whole-dollar amount in the dashboard |
| 429 Too Many Requests | Concurrency above the current limit | Respect Retry-After and lower parallelism |

Because the SDK, the wire format and the error taxonomy are identical on both endpoints, the switch is reversible at any time: point base_url back to api.anthropic.com (or delete the override) and the same code talks to Anthropic directly again. Many teams keep both clients constructed side by side during a migration week and route a small percentage of traffic to the new endpoint before flipping fully.

> Existing integrations on the legacy https://api.apitoken.sale host keep working. The unified router at router.apitoken.sale is the recommended endpoint for new setups because one base URL serves all four providers.

## Frequently asked questions

### Can I keep using the official Anthropic SDK?

Yes. Set base_url (Python) or baseURL (TypeScript) to https://router.apitoken.sale and everything else — imports, model IDs, streaming, error handling — stays the same.

### Do model IDs change when I switch base URL?

No. Use the same IDs as on the official API, such as claude-opus-4-8, claude-sonnet-5 and claude-haiku-4-5.

### Should the base URL end with /v1?

No. The SDK appends /v1/messages to whatever root you pass, so a trailing /v1 breaks the path. Pass https://router.apitoken.sale exactly.

### Do streaming and tool use work through a custom base URL?

Yes. The gateway serves the standard Anthropic Messages API, so SSE streaming, tool calling, system prompts and prompt caching behave exactly as with api.anthropic.com.

### How do I switch back to Anthropic later?

Remove the base_url / baseURL argument or unset ANTHROPIC_BASE_URL. The SDK then defaults back to https://api.anthropic.com — no other code change is needed.

---
Get a key: https://apitoken.sale/register
More guides: https://apitoken.sale/docs/learn