← Volver a las guías

Call Claude and GPT from Node.js, Fast

Why Node.js and TypeScript apps hit API friction

If you are building a chatbot, internal tool, or automation in Node.js, the biggest pain is usually not the prompt. It is the plumbing: authentication, model names, SDK compatibility, streaming, and TypeScript types. The good news is that you do not need separate client logic for Claude and GPT if you use an OpenAI-compatible relay.

59API is a low-cost relay that exposes Claude and GPT through a single endpoint at https://api.59api.com. It is designed for pay-as-you-go usage, supports Claude models like Opus, Sonnet, Haiku, and Fable, and gives you official-quality model access without the usual downgrade tradeoff. For teams watching spend, that can matter a lot more than shaving a few milliseconds off setup time.

Step 1: Install one SDK and point it at the relay

In most Node.js projects, the fastest path is to use the OpenAI SDK, even if you are calling Claude models. That is the advantage of compatibility: one client, multiple model families.

• Install the SDK: npm install openai
• Set your API key in an environment variable, for example API_KEY
• Configure the client to use https://api.59api.com as the base URL
• Choose the model you want to call, such as a GPT model or a Claude model

A typical setup in TypeScript looks like this in plain structure: create a new client, pass the key from process.env, and set baseURL to https://api.59api.com. If you keep your configuration in one file, you can switch between Claude and GPT by changing only the model string.

Step 2: Fix the most common request errors

401 Unauthorized: This usually means your key is missing, copied incorrectly, or loaded too late. In Node.js, make sure dotenv or your hosting platform loads environment variables before the client is created.

404 Model not found: This is often a model naming issue, not an SDK issue. Double-check the exact model identifier supported by the relay. For Claude, use the model name that 59API documents for the specific family you want, whether that is Opus, Sonnet, Haiku, or Fable. For GPT, use the appropriate OpenAI-style model name.

Stream never starts: If you enabled streaming, verify that your server route is not buffering the response. In Express, serverless functions, and Next.js route handlers, streaming can be interrupted if you return JSON too early or close the connection before the final chunk arrives.

TypeScript compile errors: These usually happen when your response handling is too loosely typed. Keep your client instance typed, and type the response shape you actually read. If you are only extracting text, avoid overcomplicating the object graph.

Step 3: Make TypeScript safer without making it painful

TypeScript works best here when you separate configuration, request logic, and response parsing. Use a dedicated file for your API client so that every request shares the same base URL, key, and timeout settings.

• Store the key in process.env, not in source code
• Use strict null checks so missing environment variables fail early
• Create a small wrapper function like askModel(prompt, model)
• Normalize Claude and GPT responses into one text output before returning from your route

This structure makes it easy to swap providers, test prompts, and log latency. It also helps when you are comparing cost per request across models, which is where 59API becomes attractive: you can run the same feature against cheaper Claude or GPT options and keep spend predictable.

FAQ: What developers usually ask

Can I use one code path for both Claude and GPT? Yes. If your SDK is OpenAI-compatible, keep one client and pass different model names. That is the cleanest way to avoid provider-specific conditionals all over your codebase.

Do I need separate accounts for Claude and GPT? Not necessarily. With a relay like 59API, you use one API key and one base URL, which simplifies local development and production deployment.

Is this useful for Claude Code or Codex workflows too? Yes. Since 59API is compatible with Claude Code, Codex, and OpenAI SDKs, you can reuse the same relay setup across app code and coding assistants.

Why choose 59API instead of direct provider integrations? If you want low-cost, pay-as-you-go access, a single integration surface, and access to native official-quality models, it is a very practical choice. The referral rebate is an extra bonus if you recommend it to other developers or teams.

If you are still wiring up your first Node.js or TypeScript integration, sign up for 59API and test one small request first. Once that works, expand to streaming, retries, and production logging with confidence.

Precios · Docs