Get structured output from LLMs with Anthropic

Prerequisites

Step-by-Step

1. 1

   Install the SDK

   The official Anthropic SDK handles schema conversion.

   pnpm add @anthropic-ai/sdk zod

2. 2

   Define a tool for structured extraction

   Create a tool whose input schema matches your desired output structure.

   import Anthropic from '@anthropic-ai/sdk';
   import { z } from 'zod';
   import { zodToJsonSchema } from 'zod-to-json-schema';
   
   const ExtractedData = z.object({
     name: z.string(),
     email: z.string().email(),
     company: z.string(),
     interests: z.array(z.string()),
   });
   
   const tool = {
     name: 'extract_contact',
     description: 'Extract structured contact information from text',
     input_schema: zodToJsonSchema(ExtractedData),
   };

3. 3

   Force the tool call

   Use tool_choice to make Claude always use your extraction tool.

   const client = new Anthropic();
   const message = await client.messages.create({
     model: 'claude-sonnet-4-5-20250514',
     max_tokens: 1024,
     tools: [tool],
     tool_choice: { type: 'tool', name: 'extract_contact' },
     messages: [{
       role: 'user',
       content: 'Extract contact info: John Smith, john@acme.com, works at Acme Corp, interested in AI and ML'
     }],
   });

4. 4

   Parse the tool call arguments

   The structured data is in the tool_use content block's input field.

   const toolUse = message.content.find((b) => b.type === 'tool_use');
   if (toolUse && toolUse.type === 'tool_use') {
     const data = ExtractedData.parse(toolUse.input);
     console.log(data.name, data.email);
   }

5. 5

   Add validation with Zod

   Always validate the extracted data against your schema. Claude follows schemas closely but validation adds safety.

   try {
     const validated = ExtractedData.parse(toolUse.input);
   } catch (e) {
     console.error('Extraction failed validation:', e);
   }

6. 6

   Handle complex nested structures

   Claude handles nested objects and arrays well. Keep descriptions clear for each field.

   const Order = z.object({
     customer: z.object({ name: z.string(), address: z.string() }),
     items: z.array(z.object({ product: z.string(), quantity: z.number(), price: z.number() })),
     total: z.number(),
   });

Common Pitfalls

What's Next

• ->Build extraction chains that process documents in batches.
• ->Combine with Claude's extended thinking for complex extraction tasks.
• ->Add post-processing validation for business rules beyond schema.

More Get structured output from LLMs