SSE Stream Parser for LLM APIs
Debug a streaming LLM response: paste the raw data: / event: lines (from curl, logs or the HAR analyzer) and see what the client should have reconstructed: text, tool calls, thinking and usage.
| # | Event | Summary |
|---|
How to use the SSE Stream Parser
- Capture the raw stream with
curl -Nor from a HAR file. - Paste the
data:/event:lines. - Read the reassembled text, tool calls, thinking and usage, or the raw event timeline.
Why use this SSE Stream Parser
- OpenAI Chat, OpenAI Responses, Anthropic and Gemini streams
- Reassembles fragmented tool-call JSON
- Event timeline for debugging
Related tools
Frequently Asked Questions
How do I capture a raw streaming response?
Run the request with curl -N (no buffering) and save the output, or record it in your browser's DevTools and export a HAR file; the HAR analyzer has a "Parse stream" button for event-stream responses.
Why are my tool-call arguments invalid JSON when streaming?
Arguments arrive in fragments (OpenAI function.arguments deltas, Anthropic input_json_delta). They only form valid JSON once every fragment for that call index or content block has been concatenated in order. This tool shows the reassembled result and whether it parses.
Which formats are supported?
OpenAI Chat Completions chunks, OpenAI Responses API events, Anthropic Messages streaming events (including thinking and tool use) and Gemini streamGenerateContent with alt=sse.
Is the response sent anywhere?
No. Parsing happens in your browser.