Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/phpaisdk/anthropic/llms.txt

Use this file to discover all available pages before exploring further.

Anthropic does not natively support a json_schema response format the way some other providers do. The PHP AI SDK adapts this capability by injecting a hidden structured_output tool into every request that uses .output(), then forcing Claude to call it. The tool’s input — validated against your schema — is decoded and returned as the structured result.

​
Basic Structured Output

Call .output() with a Schema::object() describing the shape you expect. Access the decoded array via $result->output or the JSON string via $result->text. 
use AiSdk\Anthropic;
use AiSdk\Generate;
use AiSdk\Schema;

$result = Generate::text()
    ->model(Anthropic::model('claude-sonnet-4'))
    ->prompt('Extract the city and country from: Lahore, Pakistan.')
    ->output(Schema::object(
        name: 'address',
        properties: [
            Schema::string(name: 'city')->required(),
            Schema::string(name: 'country')->required(),
        ],
    ))
    ->run();

// Decoded PHP array
$data = $result->output; // ['city' => 'Lahore', 'country' => 'Pakistan']

// JSON string
echo $result->text; // {"city":"Lahore","country":"Pakistan"}

​
How It Works

When .output() is present, the request builder:
  1. Constructs a hidden tool named structured_output whose input_schema is derived from your Schema::object().
  2. Appends it to body['tools'].
  3. Sets tool_choice to { "type": "tool", "name": "structured_output" }, forcing Claude to call exactly that tool.
On response, the parser detects the tool_use block with name === "structured_output", JSON-encodes its input field, and wraps it in a TextPart. This means $result->text is the JSON string and $result->output is the decoded array. The injected tool definition looks like this in the request body: 
{
  "name": "structured_output",
  "description": "Return the response as structured JSON.",
  "input_schema": {
    "type": "object",
    "properties": {
      "city":    { "type": "string" },
      "country": { "type": "string" }
    },
    "required": ["city", "country"]
  }
}

​
Combining with User Tools

When you attach both .tools() and .output() in the same request, the SDK merges them rather than replacing your tools. User-defined tools appear first, followed by structured_output. The tool_choice remains forced to structured_output. 
$weather = Tool::make('weather', 'Get weather')
    ->input(Schema::string(name: 'city')->required())
    ->run(fn (string $city): string => "Sunny in {$city}");

$result = Generate::text('Extract Lahore and keep tools available.')
    ->model(Anthropic::model('claude-sonnet-4'))
    ->tools($weather)
    ->output(Schema::object(
        name: 'address',
        properties: [Schema::string(name: 'city')->required()],
    ))
    ->run();
The tools array sent to the API will be ["weather", "structured_output"] and tool_choice will be { "type": "tool", "name": "structured_output" }.
Because tool_choice is forced to structured_output, Claude will not invoke your other tools in the same turn. Use structured output in a dedicated generation step, then perform tool calls in a separate follow-up request if needed.

​
Capability Status

This capability is reported as Adapted because the SDK implements it via forced tool use rather than a native API feature. 
use AiSdk\Capability;
use AiSdk\CapabilitySupportState;

$support = Anthropic::model('claude-sonnet-4')->capability(Capability::StructuredOutput);

$support->state;    // CapabilitySupportState::Adapted
$support->strategy; // "forced tool use"

Build docs developers (and LLMs) love

Get started for freeTalk to us