Before talking about Claude, let us understand the general idea.

API stands for application programming interface. That is a long phrase for a simple idea: a defined way for programs to talk to each other. One program sends a request, which is a structured message asking for something. The other program sends back a response, which is the answer. The rules about what a request must look like and what a response will contain are the “interface”.

You use APIs every day without noticing. A weather widget on your phone asks a weather service for today’s forecast. A travel site asks airlines for ticket prices. A payment page asks a bank to confirm a card. In each case, one program asks and another answers, using agreed rules.

Now we can apply the idea to Claude.

According to Anthropic’s documentation, the Claude API is a web-based interface, of the kind called RESTful, located at api.anthropic.com. It provides programmatic access to Claude models. Programmatic means that a program, not a person typing in a chat window, sends the requests. The main part of the API is the Messages API, which is used to send messages to Claude and receive its replies.

The documentation also describes a second general-availability API, the Message Batches API, which processes large numbers of message requests asynchronously, meaning you submit them and collect the results later, at a reduced cost. The overall set of tools for developers, including the API, the software development kits and the web console, is called the Claude Developer Platform.

This is the most common point of confusion, so let us make it clear.

The Claude app on the web, desktop and mobile is a ready-made product. You open it, type or speak, and Claude replies. It comes with features such as projects, memory, file uploads and settings, all designed for people. The Claude API is different. It is the raw doorway that developers use to build their own products. When you use a company’s customer-support chatbot or a writing tool that is powered by Claude, the company’s software is probably calling the Claude API behind the scenes.

The two are also paid for differently. The Claude app uses subscription plans with usage limits. The API is billed by how much you use it, measured in tokens, which we explain shortly. A subscription to the app does not automatically cover API usage, and API use does not require an app subscription.

Following one request from start to finish makes the whole idea concrete.

flowchart LR
  A["Your app or program"] --> B["Request with model, messages and key"]
  B --> C["Claude API"]
  C --> D["Claude model writes a reply"]
  D --> E["Response returned"]
  E --> A
        

A request is a bundle of a few simple ingredients.

Anthropic’s documentation lists the required headers for direct requests: your API key from the Console, an API version, and a content type. When Claude is accessed through a cloud platform, authentication is instead integrated with that provider’s identity system.

Tokens explain both how much text fits in a request and how you are billed.

A token is a small chunk of text, such as a short word or a piece of a longer word. Claude reads and writes in tokens. Everything you send, called input, and everything Claude writes back, called output, is measured in tokens. The total amount of material Claude can consider at once, including your messages and its reply, is limited by its context window, which is measured in tokens too.

Tokens matter for two practical reasons. First, they set the size of what you can send: a very long document may not fit. Second, they set the price: the API is billed by tokens used, rather than by seat or by subscription. Output tokens usually cost more than input tokens, and different models have different rates.

flowchart TD
  A["Your messages and instructions"] --> B["Input tokens"]
  C["Claude's reply"] --> D["Output tokens"]
  B --> E["Billed by usage"]
  D --> E
  B --> F["Counts toward the context window"]
  D --> F
        

The first steps happen in a web dashboard rather than in code.

Anthropic’s documentation says that to use the Claude API you need a Claude Console account and an API key. The Console is the web dashboard where you manage your developer account. It includes a tool called the Workbench, which lets you try the API in your browser, and a settings area where you can generate API keys. You can also use workspaces to separate your keys and control spending by project or use case.

You rarely need to build requests entirely by hand.

A client SDK, short for software development kit, is a ready-made code library that makes it easier to call an API from a particular programming language. Instead of assembling every detail of a request yourself, you use simple commands from the library. Anthropic provides client SDKs, and the documentation includes examples. Reading your key from an environment setting, rather than writing it into your code, is the pattern the documentation shows.

Anthropic’s developer documentation lives at platform.claude.com/docs, alongside the Console. An older address, docs.claude.com, redirects there, and the Console that used to be at console.anthropic.com now redirects to platform.claude.com. The documentation includes a Get started guide, a guide to working with messages, a complete API reference and release notes.

Almost anything that benefits from understanding or writing language can use Claude through the API.

The API offers building blocks that the chat app hides or packages differently.

Anthropic’s pricing materials and third-party guides describe the Batch API as giving a 50 percent reduction in token costs for work that can wait, and prompt caching as making repeated input much cheaper to read back. Exact discounts, which features are generally available and which are in beta change over time, so check the current documentation before planning around them.

flowchart TD
  A["Your task"] --> B{"Need an instant answer?"}
  B -- "Yes" --> C["Standard Messages API request"]
  B -- "No, large volume" --> D["Message Batches API at lower cost"]
  C --> E{"Same long context repeated?"}
  E -- "Yes" --> F["Consider prompt caching"]
  E -- "No" --> G["Send as normal"]
        

With the API, you choose which Claude model answers each request.

Claude comes in tiers. At the time of writing, the most recent publicly available models are Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.1, which Anthropic identifies with model strings such as claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5 and claude-fable-5-1. In general, lighter models are quicker and cheaper, and more capable models are better suited to harder tasks. Because model lineups and strings change, always copy the current names from the documentation.

Besides Anthropic’s own API, Claude is offered through some cloud providers.

Anthropic’s documentation lists several platforms through which Claude can be reached, including options on Amazon Web Services, Google Cloud and Microsoft’s Azure environment. On these, authentication is integrated with the cloud provider’s identity and access system, and feature availability can differ from the direct API. Organisations already working inside one of these clouds often choose this route for billing and security reasons.

Understanding the meter and the speed limits helps you avoid surprises.

API pricing is described as usage-based: you pay for tokens consumed rather than a monthly seat fee, with different rates for input and output and for each model. Many guides note that discounts exist for batch processing and for cached repeated input. Because prices differ across models and have changed over time, avoid relying on a number from an article without checking its date, and confirm on Anthropic’s pricing page.

The API also has rate limits, which cap how many requests and tokens you can use per minute. Guides describe a tier system in which limits rise as your account’s usage and spending grow. Workspaces let you set spending controls by project. If you exceed a limit, or something goes wrong, the API returns an error, a response that explains the problem. The documentation includes a page on errors.

An API key is like a password plus a credit card, so it deserves careful handling.

Anyone who has your API key can make requests that are billed to your account. That makes it important to keep the key secret. Do not paste it into public places, share it in chat messages or put it directly into code that others can see, such as in a public website. Documentation examples read the key from an environment setting for this reason. A common approach for websites is to keep the key on a server you control, so that visitors never see it.

These errors appear often, and each has an easy fix.

Here is the whole picture in one place.

A set of rules that lets one program ask another program for something and get an answer back, like a waiter carrying orders between a customer and a kitchen.

It is Anthropic’s programmatic doorway to Claude models. Its main part is the Messages API for sending messages to Claude and receiving replies, and there is also a Message Batches API for large volumes of asynchronous requests.

The app is a finished product for people. The API is a building block for developers who want to add Claude to their own software. They have separate accounts and separate billing.

Anthropic’s documentation says you need a Claude Console account and an API key, which you create in the Console. The Workbench lets you try things in the browser first.

A small chunk of text, such as a short word or part of a word. Claude reads and writes in tokens, and the API bills by the tokens you send and receive.

It is usage-based and depends on the model, with separate rates for input and output tokens, and discounts for certain features such as batching and caching. Prices change, so check Anthropic’s current pricing page.

Not automatically. Each request is separate. To continue a conversation, your program sends the earlier messages along with the new one, or supplies saved context itself.

Keep it out of public code and web pages, store it in environment settings or a secrets manager, use separate keys per project, set spend limits and delete or replace a key immediately if it leaks.

Yes. Anthropic’s documentation lists several cloud platforms that offer Claude, with authentication through the provider’s own system. Feature availability can differ, so check the features overview.

The developer documentation at platform.claude.com/docs, including the Get started guide, the API reference and release notes, and Anthropic’s pricing page. Details change often, so check dates.

An API is a standard way for one program to ask another for something and receive a response. The Claude API is Anthropic’s programmatic doorway to Claude models, with the Messages API at its heart and a Batches API for large asynchronous jobs. It is different from the Claude app: the app is a finished product for people, while the API is a set of building blocks for developers, accessed with a Console account and an API key, and paid for by the tokens you use rather than by subscription. Each request names a model, carries messages and settings, and returns a reply, and each request is separate unless your program supplies earlier context. You can start in the Workbench, use client SDKs and the documentation, choose a model tier to balance quality, cost and speed, and, if you prefer, reach Claude through certain cloud platforms. Keep your key secret, set spend limits, expect rate limits and check current documentation, because models, prices and features change often.