Use the Anthropic SDK

Keep the official Anthropic client and Messages API request format. Change the client’s base URL and API key, then choose a model from the Corvex Token Factory catalog. Model IDs and feature support differ from Anthropic’s hosted models.

Prerequisites

A Token Factory API key. See Authentication.
Python with the anthropic package, or Node.js with the @anthropic-ai/sdk package.
A model ID from the live catalog.

Set the key in your shell:

export CORVEX_API_KEY="sk-corvex-YOUR_VIRTUAL_KEY"

Configure the client

import os

from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["CORVEX_API_KEY"],
    base_url="https://api.tokenfactory.corvex.cloud",
)

message = client.messages.create(
    model="zai-org/GLM-5.3",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain this function."}],
)
print(next(block.text for block in message.content if block.type == "text"))

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

const client = new Anthropic({
  apiKey: process.env.CORVEX_API_KEY,
  baseURL: "https://api.tokenfactory.corvex.cloud",
});

const message = await client.messages.create({
  model: "zai-org/GLM-5.3",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Explain this function." }],
});
console.log(message.content.find((block) => block.type === "text")?.text);

The highlighted constructor fields route the client to Token Factory. The base URL has no /v1 suffix; the SDK appends /v1/messages itself and sends the key in the x-api-key header. max_tokens is required by the Messages API. Both Token Factory models reason before answering, so content starts with a thinking block followed by the text block. Select the block by type rather than by position, as the examples do. Reasoning tokens count toward max_tokens.

Run the repository examples

# Python
python -m pip install anthropic
python integrations/anthropic-drop-in/python.py

# Node.js
npm install @anthropic-ai/sdk
node integrations/anthropic-drop-in/node.mjs

A non-2xx response raises the SDK’s API error type with the Anthropic-shaped Token Factory error body. See Errors: Anthropic endpoints.

Streaming and tool use

The endpoint accepts stream: true and returns the Anthropic Server-Sent Events sequence (message_start, content_block_delta, …, message_stop), so the SDK’s streaming helpers work as they do against Anthropic. Tool definitions and tool_use / tool_result content blocks are accepted for both Token Factory models.

Count tokens

POST /v1/messages/count_tokens validates the key and request shape without running inference:

count = client.messages.count_tokens(
    model="deepseek-ai/DeepSeek-V4-Flash-0731",
    messages=[{"role": "user", "content": "Hello"}],
)
print(count.input_tokens)

Compatibility notes

Model IDs. Use the Token Factory namespace/model ID exactly as GET /v1/models returns it. Anthropic model names are not served.
Prompt caching. The endpoint accepts cache_control but Token Factory does not treat it as a request to cache content. See Errors: Prompt caching.
Input types. Neither model accepts image input. Unsupported inputs return a structured error.

Connect Claude Code — the same Messages API from the Claude Code CLI.
Use the OpenAI SDK — the Chat Completions surface for OpenAI clients.
API reference — supported Messages API request and response fields.

Get Started

Models & Pricing

Integrations

API Reference

Security & Compliance

Reference

Prerequisites

Configure the client

Run the repository examples

Streaming and tool use

Count tokens

Compatibility notes

​Prerequisites

​Configure the client

​Run the repository examples

​Streaming and tool use

​Count tokens

​Compatibility notes

​Related documentation

Prerequisites

Configure the client

Run the repository examples

Streaming and tool use

Count tokens

Compatibility notes

Related documentation