Anthropic (Claude)

Install

npm install agentfootprint @anthropic-ai/sdk

The @anthropic-ai/sdk is a peer dependency — agentfootprint’s AnthropicAdapter wraps it. Default max_tokens is 4096 — override with maxTokens in the provider options.

Server-side (Node.js)

import { Agent, anthropic, defineTool } from 'agentfootprint';

const provider = anthropic('claude-sonnet-4-20250514', {
  apiKey: process.env.ANTHROPIC_API_KEY,  // or set ANTHROPIC_API_KEY env var
});

const agent = Agent.create({ provider })
  .system('You are a helpful assistant.')
  .tool(myTool)
  .streaming(true)
  .build();

const result = await agent.run('Hello');

Browser / Edge

No SDK dependency — calls the Anthropic REST API directly:

import { Agent, BrowserAnthropicAdapter } from 'agentfootprint';

const provider = new BrowserAnthropicAdapter({
  apiKey: userApiKey,  // from your settings UI
  model: 'claude-sonnet-4-20250514',
});

const agent = Agent.create({ provider })
  .system('You are a helpful assistant.')
  .streaming(true)
  .build();

The browser adapter sends anthropic-dangerous-direct-browser-access header. This is for prototyping and playground use only. For production, proxy API calls through your backend to protect your API key.

Available models

Model	ID	Best for
Claude Sonnet 4	claude-sonnet-4-20250514	Best balance of speed and quality
Claude Opus 4	claude-opus-4-20250514	Complex reasoning, coding
Claude Haiku 3.5	claude-haiku-3-5-20241022	Fast, low cost

Extended thinking

Claude models support extended reasoning. When enabled via the Anthropic API, the model’s thinking is included in the streamed response as tokens. The full reasoning appears in the narrative trace:

const agent = Agent.create({ provider })
  .system('You are a math tutor. Show your reasoning step by step.')
  .streaming(true)
  .build();

await agent.run('If x² + 3x - 10 = 0, find x', {
  onEvent: (event) => {
    if (event.type === 'token') {
      process.stdout.write(event.content);
    }
  },
});

// The narrative captures the full LLM response including reasoning
console.log(agent.getNarrative());

Extended thinking content appears as part of the LLM response text in the narrative. The thinking event type in AgentStreamEvent is reserved for future dedicated support.

Observability

Track token usage and cost with Anthropic pricing:

import { agentObservability } from 'agentfootprint/observe';

const obs = agentObservability();

const agent = Agent.create({ provider })
  .recorder(obs)
  .build();

await agent.run('Hello');

console.log(obs.tokens());
// { totalCalls: 1, totalInputTokens: 24, totalOutputTokens: 15, calls: [...] }

// Cost is auto-calculated from model pricing
console.log(obs.cost());
// 0.000297 (USD)

Switching from the Anthropic SDK

If you’re currently using @anthropic-ai/sdk directly:

Before (raw SDK)

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

const client = new Anthropic();
const response = await client.messages.create({
  model: 'claude-sonnet-4-20250514',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello' }],
});
console.log(response.content[0].text);

After (agentfootprint)

import { Agent, anthropic } from 'agentfootprint';

const agent = Agent.create({
  provider: anthropic('claude-sonnet-4-20250514'),
})
  .system('You are helpful.')
  .build();

const result = await agent.run('Hello');
console.log(result.content);
console.log(agent.getNarrative());  // bonus: full execution trace

What you gain: tool use loop, narrative trace, grounding analysis, mock() testing, streaming events, memory, instructions — all with the same Claude models.