matts-starter-prefactor 0.1.0

Anthropic SDK wrapper with observability and tracing (Python)

Prefactor Anthropic SDK (Python)

Python wrapper for the Anthropic SDK with built-in Prefactor observability and tracing.

Installation

pip install prefactor-anthropic

Quick Start

from anthropic import Anthropic
from prefactor_anthropic import wrap_anthropic_client, shutdown

# Create your Anthropic client
client = Anthropic(api_key="your-anthropic-api-key")

# Wrap it with Prefactor observability
wrapped_client = wrap_anthropic_client(client, {
    "api_token": "your-prefactor-token",
    "agent_id": "your-agent-id"
})

# Use it like normal - all calls are automatically tracked
response = wrapped_client.messages.create(
    model="claude-sonnet-4-5-20250929",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello, Claude!"}
    ]
)

print(response.content[0].text)

# Gracefully shutdown and flush telemetry
shutdown()

Configuration

Simple Configuration

wrapped_client = wrap_anthropic_client(client, {
    "api_url": "https://api.prefactor.ai",  # Optional, defaults to env var
    "api_token": "your-token",               # Required
    "agent_id": "your-agent-id",             # Optional
    "max_input_messages": 10,                # Limit message history in spans
    "max_output_content_length": 1000,       # Truncate long outputs
})

Advanced Configuration

wrapped_client = wrap_anthropic_client(client, {
    "prefactor_config": {
        "transport_type": "http",
        "http_config": {
            "api_url": "https://api.prefactor.ai",
            "api_token": "your-token",
            "agent_id": "your-agent-id",
            "agent_identifier": "my-agent-v1",
            "agent_name": "My Custom Agent",
            "agent_description": "Custom agent for X task"
        }
    }
})

Environment Variables

You can also configure using environment variables:

export PREFACTOR_API_URL=https://api.prefactor.ai
export PREFACTOR_API_TOKEN=your-token
export PREFACTOR_AGENT_ID=your-agent-id

Then:

# Config will be loaded from environment
wrapped_client = wrap_anthropic_client(client)

Features

• ✅ Automatic tracing - All Anthropic API calls are tracked
• ✅ Token usage tracking - Input/output tokens recorded
• ✅ Streaming support - Works with streaming responses
• ✅ Error tracking - Captures and reports errors
• ✅ Configurable limits - Control message history and output length
• ✅ Type hints - Full typing support with TypedDict

Streaming Example

stream = wrapped_client.messages.stream(
    model="claude-sonnet-4-5-20250929",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Tell me a story"}]
)

for event in stream:
    if hasattr(event, 'delta') and hasattr(event.delta, 'text'):
        print(event.delta.text, end='', flush=True)

shutdown()

API Reference

wrap_anthropic_client(client, config=None)

Wraps an Anthropic client with Prefactor observability.

Parameters:

• client (Anthropic): The Anthropic client to wrap
• config (PrefactorAnthropicConfig, optional): Configuration options

Returns:

• AnthropicWrapper: Wrapped client with observability

shutdown()

Gracefully shuts down the Prefactor SDK and flushes pending telemetry data.

Development

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Type checking
mypy src/

License

MIT

Links

• Prefactor Documentation
• Anthropic SDK
• GitHub Repository