Anthropic Stream Event Viewer

Decode an Anthropic Messages SSE stream into text, thinking and tool_use blocks, stop_reason and token usage including cache writes and reads.

SSE log (14 events)

event: message_start
data: {"type":"message_start","message":{"id":"msg_01XYZ","model":"claude-opus-4-7","role":"assistant","content":[],"stop_reason":null,"usage":{"input_tokens":42,"cache_creation_input_tokens":0,"cache_read_input_tokens":1240,"output_tokens":1}}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"The user asked about caching. I should explain"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":" the difference between cache_creation and cache_read."}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"Prompt caching has two phases: write and read."}}

event: content_block_stop
data: {"type":"content_block_stop","index":1}

event: content_block_start
data: {"type":"content_block_start","index":2,"content_block":{"type":"tool_use","id":"toolu_01","name":"get_pricing","input":{}}}

event: content_block_delta
data: {"type":"content_block_delta","index":2,"delta":{"type":"input_json_delta","partial_json":"{\"model\""}}

event: content_block_delta
data: {"type":"content_block_delta","index":2,"delta":{"type":"input_json_delta","partial_json":":\"sonnet\"}"}}

event: content_block_stop
data: {"type":"content_block_stop","index":2}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"output_tokens":89}}

event: message_stop
data: {"type":"message_stop"}

Message

msg_01XYZ

Model

claude-opus-4-7

Role

assistant

Stop reason

tool_use

Model is requesting a tool call; supply tool_result and resume.

Usage

Input

Output

Cache write

Cache read

1,240

Reconstructed content blocks (3)

[0]thinking

The user asked about caching. I should explain the difference between cache_creation and cache_read.

[1]text

Prompt caching has two phases: write and read.

[2]tool_useget_pricing

accumulated input_json:

{"model":"sonnet"}

{
  "model": "sonnet"
}

Event counts

content_block_delta5

content_block_start3

content_block_stop3

message_start1

message_delta1

message_stop1

Event timeline

#	type	summary
0	message_start	model=claude-opus-4-7
1	content_block_start	idx=0 type=thinking
2	content_block_delta	idx=0 delta=thinking_delta
3	content_block_delta	idx=0 delta=thinking_delta
4	content_block_stop
5	content_block_start	idx=1 type=text
6	content_block_delta	idx=1 delta=text_delta
7	content_block_stop
8	content_block_start	idx=2 type=tool_use name=get_pricing
9	content_block_delta	idx=2 delta=input_json_delta
10	content_block_delta	idx=2 delta=input_json_delta
11	content_block_stop
12	message_delta	stop_reason=tool_use
13	message_stop

What this tool does

The Anthropic Stream Event Viewer decodes a raw message_stream from the Claude API, parsing typed events like message_start, content_block deltas, tool_use, and thinking. It reconstructs the full reply and stop_reason and breaks down per-call token usage, including cache-write and cache-read counts, so you can debug streaming and verify billing in your browser.

Updated October 2, 2026. Provided as is. Check the output before you rely on it in production.

How to use Anthropic Stream Event Viewer

1

Drop the message_stream log

Paste a raw Anthropic SSE capture — the typed event names like message_start, content_block_delta, and message_stop must remain on their own lines, with JSON payloads on data lines.
2

Read the message header

The Message panel surfaces the message id, model, role, and stop_reason as soon as those events arrive. A short note explains every stop_reason value (end_turn, tool_use, max_tokens, refusal, pause_turn).
3

Walk reconstructed content blocks

Text, thinking, and tool_use blocks are reassembled from their deltas. tool_use blocks show the accumulated partial_json and parse it once it becomes valid JSON so you can verify tool inputs end-to-end.
4

Audit usage and the timeline

The Usage panel separates input, output, cache_creation_input_tokens, and cache_read_input_tokens. The event timeline lets you click any event to see its full JSON payload for deeper debugging.

Questions and answers

Which events does it understand?

All Anthropic message_stream typed events: message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop, ping, and error. Unknown event types render as raw JSON for inspection.

How are content blocks reconstructed?

text_delta, thinking_delta, and input_json_delta payloads accumulate against the index from each content_block_start. tool_use input_json is rendered raw and re-parsed once it becomes valid JSON so you can verify tool arguments.

What does the usage panel show?

input_tokens, output_tokens, cache_creation_input_tokens, and cache_read_input_tokens — the four counters Anthropic emits. Cache write and cache read are colored differently so you can audit prompt-cache effectiveness in one glance.

Does it send my data to a server?

No. The viewer parses your SSE log, reconstructs content blocks, and renders usage entirely in your browser. Stream contents never leave your device.

How does it differ from a generic SSE replayer?

It speaks Anthropic's typed event vocabulary. Stop reasons get plain-English notes, tool_use input_json reassembles into valid JSON, and the four-channel usage breakdown is specific to Claude prompt-cache accounting.

For AI agents: how to call this tool

Machine-readable contract, endpoints and examples. Humans can ignore this section.

Best Path For Builders

Browser workflow

Runs instantly in the browser with private local processing and copy/export-ready output.

Browser Workflow

This tool is optimized for instant in-browser execution with local data handling. Run it here and copy/export the output directly.

/anthropic-stream-event-viewer/

For automation planning, fetch the canonical contract at /api/tool/anthropic-stream-event-viewer.json.

Anthropic Stream Event Viewer