Anthropic Adapter

Connects Flint to Claude models via the Anthropic Messages API. Supports streaming, tool use, vision, and prompt caching — and ships with zero runtime dependencies.

Install

npm install @flint/adapter-anthropic

Usage

import { anthropicAdapter } from '@flint/adapter-anthropic';
import { call } from 'flint';

const adapter = anthropicAdapter({
  apiKey: process.env.ANTHROPIC_API_KEY!,
});

const result = await call({
  adapter,
  model: 'claude-opus-4-5',
  messages: [{ role: 'user', content: 'Hello!' }],
});

if (result.ok) {
  console.log(result.value.message.content);
}

Constructor Options

Option	Type	Required	Description
apiKey	string	Yes	Your Anthropic API key (ANTHROPIC_API_KEY)
baseUrl	string	No	Override the API base URL (default: https://api.anthropic.com)
fetch	typeof globalThis.fetch	No	Custom fetch implementation — useful in environments without native fetch

Prompt Caching

Anthropic's prompt caching reduces cost and latency by reusing previously processed tokens. The adapter enables caching at the beta header level on every request.

To activate cache breakpoints, pass cache: 'auto' on the normalized request. When cache is 'auto', the adapter automatically places an ephemeral cache-control marker on:

• The last system message block
• The last tool definition

This covers the most common cache-worthy content. Tokens reused from cache appear in usage.cached on the response.

const result = await call({
  adapter,
  model: 'claude-opus-4-5',
  messages: [
    { role: 'system', content: longSystemPrompt },
    { role: 'user', content: 'Summarize the document.' },
  ],
  cache: 'auto',   // place ephemeral breakpoints automatically
});

if (result.ok) {
  console.log(result.value.usage.cached); // number of tokens served from cache
}

Caching is most effective for long, stable system prompts and large tool lists that remain unchanged across many calls. Cache entries have a five-minute TTL on Anthropic's side.

Capabilities

Capability	Supported
Streaming	Yes
Tool use	Yes
Parallel tools	Yes
Prompt caching	Yes
Structured output	Yes
Vision (images)	Yes
Token counting	Yes — via usage.input / usage.cached on every response

Zero Runtime Dependencies

@flint/adapter-anthropic has no npm runtime dependencies. It communicates with the Anthropic API using the platform's native fetch (or a custom implementation you supply via the fetch option), and uses types re-exported from the core flint package.

anthropicAdapter() options

function anthropicAdapter(options: {
  apiKey: string;
  baseURL?: string;       // default: https://api.anthropic.com
  defaultHeaders?: Record<string, string>;
  defaultModel?: string;  // used when model is not specified
}): ProviderAdapter

Prompt caching details

The Anthropic adapter automatically adds cache_control: { type: 'ephemeral' } to:

1. The system message (if present)
2. Tool definitions

Cache TTL is 5 minutes. Cache hits reduce input token cost by ~90%.

On cache hits, usage.cached is populated:

const res = await call({ adapter, model: 'claude-opus-4-7', messages });
if (res.ok) {
  console.log('Input tokens:', res.value.usage.input);
  console.log('Cached tokens:', res.value.usage.cached ?? 0);
}

Capabilities

adapter.capabilities = {
  streaming: true,
  toolCalling: true,
  vision: true,
  cost: true,           // reports USD cost in responses
  promptCaching: true,  // automatically managed
};

Model compatibility

All claude-* models are supported. Prompt caching is supported on claude-3-5-sonnet, claude-3-opus, and claude-opus-4-7+. Older models (claude-2, claude-instant) are not cache-aware.

Common mistakes

API key in source code

Never hardcode API keys. Use environment variables: process.env.ANTHROPIC_API_KEY!

See Also

• OpenAI-Compatible Adapter
• Writing an Adapter
• Anthropic Messages API reference
• FAQ: How does prompt caching work?
• Source: packages/adapter-anthropic/src/index.ts