Chucky SDK - Python

Python SDK for Chucky - build Claude-powered AI assistants with ease.

Installation

pip install chucky-sdk
# or
poetry add chucky-sdk
# or
uv add chucky-sdk

Quick Start

import asyncio
from chucky import Chucky

async def main():
    # Create client
    client = Chucky(
        url='wss://your-chucky-server.com/ws',
        token='your-jwt-token',
    )

    # Simple prompt
    result = await client.prompt('What is 2 + 2?')
    print(result.result)

asyncio.run(main())

Streaming

async for event in client.stream('Tell me a story'):
    if event.type == 'message':
        msg = event.data
        if msg.get('type') == 'assistant':
            content = msg.get('message', {}).get('content', [])
            for block in content:
                if block.get('type') == 'text':
                    print(block['text'], end='', flush=True)

Tools

Define tools that execute locally in your Python application:

from typing import Literal
from chucky import Chucky, tool, text_result, ToolResult

@tool("greet", "Greet someone by name")
async def greet(
    name: str,
    style: Literal["formal", "casual"] = "casual",
) -> ToolResult:
    """
    Greet someone.

    Args:
        name: The name of the person to greet
        style: The greeting style
    """
    greeting = f"Good day, {name}." if style == "formal" else f"Hey {name}!"
    return text_result(greeting)

result = await client.prompt('Greet Alice formally', tools=[greet])

MCP Servers

Group tools into MCP servers for better organization:

from chucky import create_mcp_server, tool, text_result
from datetime import datetime

@tool("get_time", "Get current time")
async def get_time() -> ToolResult:
    return text_result(datetime.now().isoformat())

time_server = create_mcp_server("time-tools", [get_time])

client = Chucky(
    url='wss://...',
    token='...',
    mcp_servers={'time-tools': time_server},
)

Tool Schema Options

The @tool decorator supports multiple ways to define input schemas:

1. Auto-infer from function signature (recommended)

@tool("greet", "Greet someone")
async def greet(name: str, count: int = 1) -> ToolResult:
    return text_result(f"Hello, {name}!" * count)

2. Pydantic model

from pydantic import BaseModel

class GreetInput(BaseModel):
    name: str
    style: Literal["formal", "casual"] = "casual"

@tool("greet", "Greet someone", schema=GreetInput)
async def greet(args: dict) -> ToolResult:
    return text_result(f"Hello, {args['name']}!")

3. Raw JSON Schema

@tool("greet", "Greet someone", schema={
    "type": "object",
    "properties": {"name": {"type": "string"}},
    "required": ["name"]
})
async def greet(args: dict) -> ToolResult:
    return text_result(f"Hello, {args['name']}!")

API Reference

Chucky

Main client class.

Constructor Parameters:

• url (required): WebSocket URL of the Chucky server
• token (required): JWT billing token
• model: Claude model to use (default: claude-sonnet-4-5-20250929)
• system_prompt: Default system prompt
• max_turns: Maximum conversation turns
• allowed_tools: List of allowed tool names
• disallowed_tools: List of disallowed tool names
• mcp_servers: MCP servers with client-side tools
• timeout: Connection timeout in seconds (default: 30.0)
• keepalive_interval: Keep-alive interval in seconds (default: 60.0)

Methods:

• prompt(message, **options): Send a prompt and wait for result
• stream(message, options=None): Send a prompt and stream events

@tool(name, description, schema=None)

Decorator to create a tool from a function.

create_mcp_server(name, tools, version="1.0.0")

Create an MCP server with tools.

text_result(text)

Helper to create a simple text result.

error_result(message)

Helper to create an error result.

License

MIT