Anthropic

Anthropic's API is a single endpoint — /v1/messages — with a rich typed streaming lifecycle. VidaiMock replicates the complete wire contract, including the 2025+ cost-accounting fields and strict event ordering that the official SDK asserts on.

Messages — POST /v1/messages

The bundled anthropic/message.json.j2 template branches on the request:

• tools present → tool_use content block, stop_reason: "tool_use", with caller.type and the caller's declared tool name.
• tools + prior tool_result in history → text content, stop_reason: "end_turn" (see Agentic testing).
• default → text content, stop_reason: "end_turn".

curl http://localhost:8100/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-4-5-20251001", "max_tokens": 200,
       "messages": [{"role": "user", "content": "Hi"}]}'

Full usage shape

Every response carries the 2025+ accounting fields real Anthropic returns:

"usage": {
  "input_tokens": 16,
  "cache_creation_input_tokens": 0,
  "cache_read_input_tokens": 0,
  "cache_creation": { "ephemeral_5m_input_tokens": 0, "ephemeral_1h_input_tokens": 0 },
  "output_tokens": 10,
  "service_tier": "standard",
  "inference_geo": "not_available"
}

plus a top-level stop_details: null.

Request validation

A /v1/messages request missing a required field returns HTTP 400 with the real Anthropic error envelope and a per-field message:

curl http://localhost:8100/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model": "claude", "messages": [{"role": "user", "content": "Hi"}]}'

{"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: Field required"}}

This matches live Anthropic, so client code that catches anthropic.BadRequestError can be tested in mock mode.

Streaming — all 7 event types

"stream": true emits the complete Anthropic SSE lifecycle in strict order:

event: message_start
event: content_block_start
event: ping
event: content_block_delta   (one or more)
event: content_block_stop
event: message_delta          (final stop_reason + usage)
event: message_stop

Events are blank-line separated so SDK parsers attribute each data: to the correct event:. Tool-mode streaming emits a tool_use content block with input_json_delta deltas and stop_reason: "tool_use" in message_delta.

curl -N http://localhost:8100/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-4-5-20251001", "max_tokens": 200,
       "stream": true, "messages": [{"role": "user", "content": "Count to 5"}]}'

Tool calling

curl http://localhost:8100/v1/messages \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-4-5-20251001", "max_tokens": 500,
       "messages": [{"role": "user", "content": "Weather in London?"}],
       "tools": [{"name": "get_weather", "description": "Get weather",
                  "input_schema": {"type": "object",
                  "properties": {"city": {"type": "string"}}}}]}'

Error envelopes

Chaos/override errors render Anthropic's {"type":"error","error":{"type","message"}} shape, with error.type selected per status (authentication_error, rate_limit_error, overloaded_error, api_error, …).