Claude protocol: Anthropic Messages API

POST https://api.laozhang.ai/v1/messages

​

Choose a model

• Claude models: 11 of them, such as claude-sonnet-5-5 and claude-opus-5-5. See the Anthropic section of the catalog for the full list and prices.
• DeepSeek: deepseek-v4-pro and deepseek-v4-flash.
• GLM: glm-5.2, used in the examples on this page, plus glm-5.3 and others. It behaves much like the official Messages API through this protocol.
• Kimi: kimi-k3, kimi-k2.6, and others.

​

Claude model limits

• Output length: each response is capped at 32,768 tokens. A larger max_tokens is reduced to the cap.
• PDFs and documents: document content blocks are not supported. Extract the text first and send it in a text block.
• Server tools: Anthropic server tools such as web_search, web_fetch, and code_execution are not supported, and requests that declare them return 400.
• Other endpoints: /v1/messages/count_tokens, Message Batches, and the Files API are not available.
• Stop sequences: when a stop sequence matches, stop_reason is end_turn and stop_sequence is null.
• Long Haiku prompts: when a claude-haiku-5-5 prompt exceeds 100,000 tokens, counting cache reads and writes, the whole request is billed at the long-context rate shown in the catalog’s tiered pricing.

​

Send a message

export LAOZHANG_API_KEY="YOUR_LAOZHANG_API_KEY"

curl --fail-with-body --max-time 120 \
  "https://api.laozhang.ai/v1/messages" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "glm-5.2",
    "max_tokens": 1024,
    "system": "You are a concise assistant.",
    "messages": [
      {"role": "user", "content": "Reply only with: connected"}
    ]
  }'

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["LAOZHANG_API_KEY"],
    base_url="https://api.laozhang.ai",
    timeout=120.0,
    max_retries=0,
)
message = client.messages.create(
    model="glm-5.2",
    max_tokens=1024,
    system="You are a concise assistant.",
    messages=[{"role": "user", "content": "Reply only with: connected"}],
)
for block in message.content:
    if block.type == "text":
        print(block.text)

​

Read the response

{
  "type": "message",
  "role": "assistant",
  "model": "glm-5.2",
  "content": [
    {"type": "text", "text": "connected"}
  ],
  "stop_reason": "end_turn",
  "usage": {"input_tokens": 21, "output_tokens": 4}
}

• end_turn: the answer is finished.
• max_tokens: the output limit cut it short.
• tool_use: the model is waiting for a tool result.

​

Stream the response

curl --fail-with-body -N --max-time 300 \
  "https://api.laozhang.ai/v1/messages" \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "glm-5.2",
    "max_tokens": 1024,
    "stream": true,
    "messages": [
      {"role": "user", "content": "Give three short tips for naming variables."}
    ]
  }'

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["LAOZHANG_API_KEY"],
    base_url="https://api.laozhang.ai",
    timeout=300.0,
    max_retries=0,
)
with client.messages.stream(
    model="glm-5.2",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Give three short tips for naming variables."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
print()

event: message_start
data: {"type":"message_start","message":{"type":"message","role":"assistant","content":[]}}

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

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"1. Name variables"}}

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

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"}}

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

• Join the text_delta pieces for each block index to rebuild the reply.
• Thinking and tool input use other delta types, and a stream can also contain ping events, so skip event types you do not handle.
• In Python, text_stream yields only the text and ends after message_stop.
• If the connection closes before message_stop, treat the reply as incomplete.

​

Use tools

1. Read the id, name, and input of each tool_use block.
2. Run the matching function in your code.
3. Send the next request with the assistant turn unchanged, followed by a user turn that holds one tool_result block per call. Each tool_use_id must match the id of the call it answers.

[
  {"role": "user", "content": "What is the weather in Paris?"},
  {"role": "assistant", "content": [
    {"type": "tool_use", "id": "toolu_01", "name": "get_weather", "input": {"city": "Paris"}}
  ]},
  {"role": "user", "content": [
    {"type": "tool_result", "tool_use_id": "toolu_01", "content": "18°C and cloudy"}
  ]}
]

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["LAOZHANG_API_KEY"],
    base_url="https://api.laozhang.ai",
    timeout=120.0,
    max_retries=0,
)
tools = [{
    "name": "get_weather",
    "description": "Get the current weather for a city",
    "input_schema": {
        "type": "object",
        "properties": {"city": {"type": "string", "description": "City name"}},
        "required": ["city"],
    },
}]


def get_weather(city):
    # Replace with your own logic, such as a call to a weather service
    return f"{city}: cloudy, 18°C"


messages = [{"role": "user", "content": "What is the weather in Paris?"}]
for _ in range(5):
    message = client.messages.create(
        model="glm-5.2",
        max_tokens=1024,
        tools=tools,
        messages=messages,
    )
    tool_uses = [block for block in message.content if block.type == "tool_use"]
    if not tool_uses:
        break
    messages.append({"role": "assistant", "content": message.content})
    results = []
    for block in tool_uses:
        if block.name == "get_weather":
            output = get_weather(**block.input)
        else:
            output = f"Unknown tool: {block.name}"
        results.append({
            "type": "tool_result",
            "tool_use_id": block.id,
            "content": output,
        })
    messages.append({"role": "user", "content": results})

print("".join(block.text for block in message.content if block.type == "text"))

​

Use with Claude Code

• Set ANTHROPIC_BASE_URL to https://api.laozhang.ai (the host root, without /v1).
• Choose a model whose Endpoints include Anthropic Messages, such as claude-sonnet-5-5 or glm-5.2.
• Use a token in the Claude Code group with the Usage-first billing mode. If Claude Code currently runs on a default-group token, create a new Claude Code-group token or edit the existing token and change its group.

​

Troubleshooting

• 401: check the Authorization: Bearer header and use a model API key from Token management, not an account AccessToken.
• 400 that mentions max_tokens: add max_tokens. Messages requests require it, unlike Chat Completions.
• The request URL contains /v1/v1/messages: the SDK base URL ends in /v1. Set it to https://api.laozhang.ai.
• 404 or model not found: check the exact model ID and confirm in the model catalog that it lists Anthropic Messages for your token group.
• 503 Current group claude_code has no available channels for model …: the model isn’t served over the Claude protocol, for example a GPT or Gemini model. Call it with a default-group token in the OpenAI-compatible format.
• 503 Current group default has no available channels for model …: default-group tokens can’t call DeepSeek, GLM, or Kimi over the Claude protocol. Use a Claude Code-group token instead. Claude models aren’t affected.
• Claude Code returns 400 ValidationException or Extra inputs are not permitted: follow Fix Claude Code 400 ValidationException errors.
• 400 after declaring web_search or another server tool on a Claude model: remove the server tool from tools. Run searches in your application and return the results as tool_result blocks.
• 529 with an upstream-busy message: capacity is short at peak times. Retry after a short wait; the Anthropic SDKs retry 529 automatically by default.

​

Related documentation

• Anthropic Messages API reference
• Anthropic streaming reference
• Anthropic tool use overview
• Anthropic model overview
• OpenAI protocol: Chat Completions
• Model availability and permissions