Skip to main content
Set stream: true to receive Anthropic-style event streams. Use this when your client already expects message_start, content_block_delta, and message_stop events.

​
Enable Streaming

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.naga.ac",
    api_key="YOUR_API_KEY",
)

with client.messages.stream(
    model="claude-sonnet-4.5",
    max_tokens=256,
    messages=[
        {"role": "user", "content": "Stream a short explanation of retries."}
    ],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

​
Event Lifecycle

Per the protocol tests, a normal text stream begins with message_start and ends with message_stop. 
event: message_start
data: {...}

event: content_block_start
data: {...}

event: content_block_delta
data: {"delta":{"type":"text_delta","text":"Hello"}}

event: message_delta
data: {"delta":{"stop_reason":"end_turn","stop_sequence":null}}

event: message_stop
data: {...}

​
Tool Streaming

Tool argument deltas appear as content_block_delta with type: input_json_delta.

​
Thinking Streaming

Thinking-related data can appear before text blocks, including signature_delta events.

​
What to listen for

  • message_start for the initial message frame and input usage
  • content_block_start and content_block_delta for text, tool, and thinking blocks
  • message_delta for stop reason and final output usage
  • message_stop for stream completion

​
Common mistakes

  • assuming every content_block_delta is text
  • ignoring message_delta and losing final usage information
  • treating Messages streams like chat chunks or Responses semantic events