1. 遇到问题
Python 基于 Langchain 对接 Claude 3.7 大模型接口进行问答时，由于国内不在Claude支持的国家和地区，所以一直调不通，错误 anthropic.PermissionDeniedError: Error code: 403 - {'error': {'type': 'forbidden', 'message': 'Request not allowed'}}
详细堆栈如下：
ERROR:root:Error code: 403 - {'error': {'type': 'forbidden', 'message': 'Request not allowed'}}
Traceback (most recent call last):File "E:\Idea_workspace\Temp_view\test-python\test\llm\test_claude37.py", line 58, in <module>for chunk in llm.stream(messages):File "D:\Program Files\Python311\Lib\site-packages\langchain_core\language_models\chat_models.py", line 494, in streamfor chunk in self._stream(input_messages, stop=stop, **kwargs):File "D:\Program Files\Python311\Lib\site-packages\langchain_anthropic\chat_models.py", line 995, in _streamstream = self._client.messages.create(**payload)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\Program Files\Python311\Lib\site-packages\anthropic\_utils\_utils.py", line 275, in wrapperreturn func(*args, **kwargs)^^^^^^^^^^^^^^^^^^^^^File "D:\Program Files\Python311\Lib\site-packages\anthropic\resources\messages\messages.py", line 953, in createreturn self._post(^^^^^^^^^^^File "D:\Program Files\Python311\Lib\site-packages\anthropic\_base_client.py", line 1336, in postreturn cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\Program Files\Python311\Lib\site-packages\anthropic\_base_client.py", line 1013, in requestreturn self._request(^^^^^^^^^^^^^^File "D:\Program Files\Python311\Lib\site-packages\anthropic\_base_client.py", line 1117, in _requestraise self._make_status_error_from_response(err.response) from None
anthropic.PermissionDeniedError: Error code: 403 - {'error': {'type': 'forbidden', 'message': 'Request not allowed'}}2. 解决问题
调用Claude 3.7 进行问答时，通常会采用流式输出 或 异步调用，这两种不同的调用 解决403有点差异。 原理都是通过设置代理客户端（如下图，配上ip:端口就能访问外网的那种）、转发请求。
代码里有注释，所以就不过多说明了，有问题可留言沟通。
from langchain_anthropic import ChatAnthropic
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
import httpx
import os
import dotenv
import logging
import asynciodotenv.load_dotenv()
INPUT_PLACEHOLDER = '{input}'
# 读取环境变量值，或直接改为你的代理地址，比如：proxy_url = http://192.168.1.101:10082
proxy_url = os.getenv("CLAUDE_PROXY")if proxy_url:#新建异步httpx客户端async_transport = httpx.AsyncHTTPTransport(proxy=httpx.Proxy(url=proxy_url))async_http_client = httpx.AsyncClient(transport=async_transport)#新建httpx客户端transport = httpx.HTTPTransport(proxy=httpx.Proxy(url=proxy_url))http_client = httpx.Client(transport=transport)
else:async_http_client = Nonehttp_client = None
llm = ChatAnthropic(thinking={"type": "enabled","budget_tokens": 4096},model="claude-3-7-sonnet-20250219",max_tokens=8192
)
if http_client:# 设置同步httpx客户端llm._client._client = http_client
if async_http_client:# 设置异步httpx客户端llm._async_client._client = async_http_client
system_prompt = """
你是一个人工智能助手，你可以回答用户的问题。"""
prompt = ChatPromptTemplate.from_messages([("system", system_prompt),("user", INPUT_PLACEHOLDER)
])
output_parser = StrOutputParser()
messages = prompt.format(input="你是谁？中文回答")# Stream 流式输出 同步调用
try:for chunk in llm.stream(messages):print("" + chunk.text(), end="")
except Exception as e:logging.exception(e)#Async 异步调用
async def main():result = await llm.ainvoke(messages)print("\n\n异步：" + result.text())asyncio.run(main())
效果如图
3. 总结
其实Claude 3.7 403问题一直困扰研发团队，不得不 花钱租用Aws服务器在上面做调试&开发，成本又高又不方便(尤其是多人使用)。
后来找时间看了些 langchain_anthropic和httpx的源码 该问题才算完美解决。
