Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/anthropics/claude-agent-sdk-python/llms.txt

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

The Claude Agent SDK provides two main ways to interact with Claude: the query() function for simple one-shot queries and ClaudeSDKClient for interactive, stateful conversations.

​
Quick Comparison

Featurequery()ClaudeSDKClient
CommunicationUnidirectionalBidirectional
StateStatelessStateful
Follow-upsNot supportedSupported
InterruptsNot supportedSupported
Connection ManagementAutomaticManual (context manager)
Use CaseSimple automationInteractive conversations

​
query() Function

The query() function is ideal for simple, stateless queries where you don’t need bidirectional communication or conversation management.

​
Signature

async def query(
    *,
    prompt: str | AsyncIterable[dict[str, Any]],
    options: ClaudeAgentOptions | None = None,
    transport: Transport | None = None,
) -> AsyncIterator[Message]

​
Key Characteristics

Unidirectional

Send all messages upfront, receive all responses. No follow-up messages can be sent.

Stateless

Each query is independent with no conversation state maintained between calls.

Simple

Fire-and-forget style with automatic connection management.

No Interrupts

Cannot interrupt or send messages during execution.

​
When to Use query()

Perfect for simple questions where you know the entire input upfront.
async for message in query(prompt="What is 2+2?"):
    if isinstance(message, AssistantMessage):
        for block in message.content:
            if isinstance(block, TextBlock):
                print(block.text)
Ideal for processing multiple independent prompts.
prompts = ["Explain async/await", "What is a closure?", "Define recursion"]

for prompt in prompts:
    async for message in query(prompt=prompt):
        # Process each response
        pass
Great for automated scripts where all inputs are known.
async for message in query(
    prompt="Review this pull request for security issues",
    options=ClaudeAgentOptions(
        cwd="/path/to/repo",
        permission_mode="default"
    )
):
    # Process review results
    pass
Useful for one-shot code generation tasks.
async for message in query(
    prompt="Create a Python web server with FastAPI",
    options=ClaudeAgentOptions(
        system_prompt="You are an expert Python developer",
        cwd="/home/user/project"
    )
):
    # Process generated code
    pass

​
ClaudeSDKClient

The ClaudeSDKClient provides full control over the conversation flow with support for streaming, interrupts, and dynamic message sending.

​
Signature

class ClaudeSDKClient:
    def __init__(
        self,
        options: ClaudeAgentOptions | None = None,
        transport: Transport | None = None,
    )

​
Key Characteristics

Bidirectional

Send and receive messages at any time during the conversation.

Stateful

Maintains conversation context across multiple message exchanges.

Interactive

Send follow-ups based on Claude’s responses in real-time.

Control Flow

Support for interrupts, permission changes, and session management.

​
When to Use ClaudeSDKClient

Building conversational UIs or chat applications.
async with ClaudeSDKClient() as client:
    # Send initial query
    await client.query("Help me debug this code")
    
    async for msg in client.receive_response():
        if isinstance(msg, AssistantMessage):
            display_to_user(msg)
    
    # User can send follow-ups
    await client.query("What about the edge case?")
    async for msg in client.receive_response():
        display_to_user(msg)
When you need to react to Claude’s responses and ask follow-up questions.
async with ClaudeSDKClient() as client:
    await client.query("Analyze this error")
    
    async for msg in client.receive_response():
        if isinstance(msg, AssistantMessage):
            # Analyze response and ask follow-up
            if "needs more context" in extract_text(msg):
                await client.query("Here's the stack trace...")
                break
When you need to maintain context across multiple exchanges.
async with ClaudeSDKClient() as client:
    await client.query("Let's refactor this module")
    async for msg in client.receive_response():
        pass  # Process initial response
    
    await client.query("Now add error handling")
    async for msg in client.receive_response():
        pass  # Claude remembers the refactoring context
When you need to change settings or interrupt during execution.
async with ClaudeSDKClient() as client:
    await client.query("Review this codebase")
    
    # Start listening for messages
    async for msg in client.receive_messages():
        if user_approves():
            # Switch permission mode dynamically
            await client.set_permission_mode('acceptEdits')
            await client.query("Implement the changes")
        elif user_cancels():
            # Interrupt the current operation
            await client.interrupt()
            break

​
Complete Examples

​
Using query() for Code Generation

from claude_agent_sdk import query, ClaudeAgentOptions

async def generate_code():
    """Generate a Python web server."""
    async for message in query(
        prompt="Create a FastAPI server with health check endpoint",
        options=ClaudeAgentOptions(
            system_prompt="You are an expert Python developer",
            cwd="/home/user/project",
            permission_mode="acceptEdits"
        )
    ):
        if isinstance(message, AssistantMessage):
            for block in message.content:
                if isinstance(block, TextBlock):
                    print(block.text)
        elif isinstance(message, ResultMessage):
            print(f"Cost: ${message.total_cost_usd:.4f}")

​
Using ClaudeSDKClient for Interactive Chat

from claude_agent_sdk import ClaudeSDKClient, ClaudeAgentOptions

async def interactive_chat():
    """Interactive debugging session."""
    options = ClaudeAgentOptions(
        cwd="/home/user/project",
        permission_mode="default"
    )
    
    async with ClaudeSDKClient(options) as client:
        # Initial query
        await client.query("Help me understand this error")
        
        async for msg in client.receive_response():
            if isinstance(msg, AssistantMessage):
                for block in msg.content:
                    if isinstance(block, TextBlock):
                        print(f"Claude: {block.text}")
        
        # Follow-up based on response
        await client.query("Can you show me how to fix it?")
        
        async for msg in client.receive_response():
            if isinstance(msg, AssistantMessage):
                # User reviews the solution
                if should_accept():
                    await client.set_permission_mode('acceptEdits')
                    await client.query("Please implement the fix")
            elif isinstance(msg, ResultMessage):
                print(f"Session complete. Cost: ${msg.total_cost_usd:.4f}")
Important Limitation: As of v0.0.20, ClaudeSDKClient instances cannot be used across different async runtime contexts (e.g., different trio nurseries or asyncio task groups). The client must complete all operations within the same async context where it was connected.

​
Decision Guide

Use query() when:
  • You know all inputs upfront
  • No follow-up messages needed
  • Automating batch operations
  • Building CI/CD integrations
  • Simple one-off tasks
Use ClaudeSDKClient when:
  • Building chat interfaces
  • Need interactive debugging
  • Require follow-up questions
  • Need to interrupt operations
  • Want to change settings mid-conversation
  • Building REPL-like tools

Build docs developers (and LLMs) love

Get started for freeTalk to us