@prompty/anthropic

Anthropic provider for Prompty — executor and processor for the Anthropic Messages API.

Installation

npm install @prompty/core @prompty/anthropic @anthropic-ai/sdk

Usage

Import @prompty/anthropic to auto-register the anthropic provider, then use @prompty/core as normal:

import "@prompty/anthropic";
import { run } from "@prompty/core";

const result = await run("./chat.prompty", {
  question: "What is quantum computing?",
});

.prompty Configuration

Set provider: anthropic in your .prompty file:

---
name: my-prompt
model:
  id: claude-sonnet-4-20250514
  provider: anthropic
  apiType: chat
  connection:
    kind: key
    apiKey: ${env:ANTHROPIC_API_KEY}
  options:
    temperature: 0.7
    maxOutputTokens: 1000
---
system:
You are a helpful assistant.

user:
{{question}}

Supported API Types

| apiType | Description | |-----------|-------------| | chat (default) | Messages via client.messages.create() |

Streaming

Enable streaming via additionalProperties:

model:
  options:
    additionalProperties:
      stream: true

Returns a PromptyStream that can be iterated asynchronously.

Structured Output

Anthropic supports structured output via outputs in the .prompty frontmatter. The executor converts this to a JSON tool-use pattern, and the processor automatically parses the result.

Exports

| Export | Description | |--------|-------------| | AnthropicExecutor | Executor implementation for Anthropic | | AnthropicProcessor | Processor for Anthropic responses | | processResponse | Shared response processing helper | | messageToWire | Convert Message → Anthropic wire format | | buildChatArgs | Build messages API arguments | | toolsToWire | Convert tools → Anthropic tool format | | outputsToWire | Convert outputs → Anthropic tool format |

License

MIT