Claude API checker 中文

Claude API checker

Enter the base URL, API key and model of an API that speaks the Anthropic Messages API. The check sends a set of probe requests and looks at whether the model behind the API is really Claude, whether features such as PDF input and tool use work, and whether responses follow the official API format.

Only the Anthropic Messages protocol (POST /v1/messages) is supported. OpenAI-format APIs can't be checked.

The root of the API; requests go to its /v1/messages. https only.

Used only for this check. It is not stored or logged, and the report shows it masked. The probe requests are billed to this key as normal usage.

The list has common models; you can also type any model name the API supports.

Each IP can run 3 checks per 10 minutes and 20 per day, one at a time.

What is checked

Authenticity

  • Identity: Asks the model who it is and checks that the answer names Claude and Anthropic.
  • Behavioral signature: Three prompts that check formatting and refusal style typical of Claude.
  • Thinking signature: Checks the signature on the thinking block. Anthropic's servers create it and a relay cannot forge it.
  • Model consistency: The model field in responses, and how much the output length varies over three runs.
  • Knowledge: Five questions about Anthropic and Claude.

Capability

  • PDF reading: Sends a PDF and checks that the model reads a code word from it.
  • Structured output: A tool call against a JSON schema, including tool call IDs and fields.

Protocol

  • Protocol: Response fields, content block types and the order of streaming events.
  • Response integrity: Compares streaming and non-streaming responses to the same prompt.
  • Message IDs: The prefixes of message and tool call IDs (msg_, toolu_).
  • Token usage: Whether the usage fields are present and add up, including against count_tokens.