streaming

Stream Claude API responses progressively via Server-Sent Events with delta types.

18|2|Updated Dec 12, 2025
One-click install
npx skills add https://github.com/Lobbi-Docs/claude --skill streaming-lobbi-docs
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: streaming
Source: https://github.com/Lobbi-Docs/claude/tree/main/.claude/skills/streaming
Command: npx skills add https://github.com/Lobbi-Docs/claude --skill streaming-lobbi-docs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires llm-integration.

What problem does it solve?

This Skill enables real-time, progressive delivery of AI responses from Claude, improving user experience for long-running generations, tool use, and extended thinking by displaying output as it's generated.

Core Features & Use Cases

  • Real-time Output: Stream text, tool use, and extended thinking content as it becomes available.
  • Event-Driven Processing: Handle different SSE event types (content_block_delta, message_delta) for granular control.
  • Robust Error Handling: Implement retry mechanisms with exponential backoff for API resilience.
  • Use Case: You're building a chatbot that generates complex reports. Use this skill to stream the report content word-by-word to the user interface, providing immediate feedback and a more engaging experience, rather than waiting for the full response.

Quick Start

Use the streaming skill to generate a short story and display the output in real-time.

Frequently Asked Questions about streaming

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I stream real-time responses from the Claude API to a user interface?

Stream real-time responses using Server-Sent Events (SSE) to progressively deliver Claude's output as it's generated. This enables word-by-word text delivery, live tool-use execution, and extended thinking visualization, improving user engagement without waiting for complete message generation.

What are the different SSE event types I need to handle for streaming?

Handle three main delta types: text_delta for streamed text content, input_json_delta for tool parameters, and thinking_delta for extended thinking output. Each event type requires distinct buffering and parsing logic to safely reconstruct content blocks without interrupting JSON structures mid-stream.

Can I use streaming with tool use in Claude API integrations?

Yes. Streaming supports tool-use workflows by delivering tool calls progressively and buffering input_json_delta events until complete, preventing JSON parsing errors. This enables real-time visibility into multi-step tool-assisted tasks without blocking on full response generation.

How do I handle streaming errors and ensure reliable API delivery?

Implement retry mechanisms with exponential backoff to handle transient failures. Gracefully manage connection drops and partial messages by validating event structures and maintaining state across reconnections for robust real-time delivery.

What client languages does Claude API streaming support?

Streaming via SSE supports Python and TypeScript clients, enabling cross-platform real-time output delivery. Both language implementations handle event parsing, delta reconstruction, and asynchronous processing patterns natively.

When should I use streaming instead of waiting for complete responses?

Use streaming for long-running generations, complex report generation, interactive chatbots, and extended thinking workflows where immediate user feedback improves perceived performance. Avoid streaming for simple, fast responses where latency is negligible.