Using Anthropic's API

Following Anthropic's getting started documentation let's create a simple code just to use the API. This is the basis for a future agent we can create.

Let's do this in Node.js. If you don't have Node installed on your system, then the node --version command will show the installed version.

❯ node --version
v22.16.0

Create any folder and inside it run the command below. This will install the libraries to be used in this folder. They will be inside the node_modules folder.

❯ npm install @anthropic-ai/sdk

Checking.

❯ tree -L 1
.
├── node_modules
├── package-lock.json
├── package.json

Now we need a main file that calls this library. Create a main.js file with the following content.

import Anthropic from "@anthropic-ai/sdk"; // Importing the library

// Initializing the library
const anthropic = new Anthropic();

// Actually this function does nothing more than build the post request that will be the call to the api.

// By default this lib looks for the ANTHROPIC_API_KEY environment variable that should be available in your terminal.
const msg = await anthropic.messages.create({
  model: "claude-opus-4-20250514",
  max_tokens: 1000,
  temperature: 1,
  // Here we declare the system prompt
  system: "Respond objectively and with short sentences",
  messages: [
    {
      role: "user",
      content: [
        {
          type: "text",
          // Here we declare the user's text.
          text: "Why is the ocean salty?"
        }
      ]
    }
  ]
});
console.log(msg);

❯ tree -L 1
.
├── main.js
├── node_modules
├── package-lock.json
├── package.json

Export the environment variable with the ANTHROPIC_API_KEY and run main.js.

To be able to make the calls you'll need to buy some credits there at Anthropic and generate an APIKEY.

export ANTHROPIC_API_KEY=xxxxxxxxxx

❯ node main.js
## OUR RESPONSE.
{
  id: 'msg_01SHqZPvKhbhLHzqqAbH2kJw',
  type: 'message',
  role: 'assistant',
  model: 'claude-opus-4-20250514',
  content: [
    {
      type: 'text',
      text: 'The ocean is salty because:\n' +
        '\n' +
        '1. **Rivers carry salts**: Rainwater dissolves minerals from rocks and rivers carry these salts to the sea\n' +
        '\n' +
        '2. **Evaporation concentrates salt**: Water evaporates, but salt remains in the ocean, increasing its concentration over millions of years\n' +
        '\n' +
        '3. **Volcanic activity**: Submarine volcanoes release minerals and salts directly into the water\n' +
        '\n' +
        '4. **Geological time**: This process has been happening for billions of years, accumulating more and more salt\n' +
        '\n' +
        'The main salt is sodium chloride (table salt), but there are other dissolved minerals too.'
    }
  ],
  stop_reason: 'end_turn',
  stop_sequence: null,
  usage: {
    input_tokens: 30,
    cache_creation_input_tokens: 0,
    cache_read_input_tokens: 0,
    output_tokens: 182,
    service_tier: 'standard'
  }
}

Notice how the response text came formatted in markdown inside a string and the use of 30 input tokens and 182 output.

If this text were exposed in markdown it would be this.

The ocean is salty because:

1. Rivers carry salts: Rainwater dissolves minerals from rocks and rivers carry these salts to the sea
2. Evaporation concentrates salt: Water evaporates, but salt remains in the ocean, increasing its concentration over millions of years
3. Volcanic activity: Submarine volcanoes release minerals and salts directly into the water
4. Geological time: This process has been happening for billions of years, accumulating more and more salt The main salt is sodium chloride (table salt), but there are other dissolved minerals too. .

The same request we're building here can be done through the console directly on Anthropic's site at https://console.anthropic.com by clicking Create Prompt

We can make a prediction with the token counter before sending the request for processing. It's actually an estimate, but there can be very small differences. Let's put it in a new file counttoken.js.

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

const client = new Anthropic();

const response = await client.messages.countTokens({
  model: 'claude-opus-4-20250514',
  system: "Respond objectively and with short sentences",
  messages: [
    {
      role: "user",
      content: [
        {
          type: "text",
          // Here we declare the user's text.
          text: "Why is the ocean salty?"
        }
      ]
    }
  ]
});

console.log(response);

Running we have the same expected value we had before.

❯ node counttoken.js
{ input_tokens: 30 }