httpx transport for curl_cffi - python binding for curl-impersonate fork

Unlike other pure python http clients like httpx (with native transport) or requests, curl_cffi can impersonate browser's TLS/JA3 and HTTP/2 fingerprints.

Browser simulation implemented by low-level customizations and usage of native browser TLS libraries (BoringSSL for Chrome, nss for Firefox) - which is impossible to achieve with Python's OpenSSL binding.

If you are blocked by some website for no obvious reason, you can give curl_cffi a try.

pip install httpx-curl-cffi

from httpx import Client, AsyncClient
from httpx_curl_cffi import CurlTransport, AsyncCurlTransport, CurlOpt

client = Client(transport=CurlTransport(impersonate="chrome", default_headers=True))
client.get("https://tools.scrapfly.io/api/fp/ja3")

async_client = AsyncClient(transport=AsyncCurlTransport(
  impersonate="chrome",
  default_headers=True,
  # required for parallel requests, see curl_cffi issues below
  curl_options={CurlOpt.FRESH_CONNECT: True}
))

Note that httpx.Client and httpx.AsyncClient disables proxy configuration from environment variables on providing transport argument even with trust_env=True (default), to have this configured in the same way as native transport use snippet below:

import httpx
from httpx._utils import get_environment_proxies
from httpx_curl_cffi import CurlTransport  # or AsyncCurlTransport

def transport_factory(proxy: httpx.Proxy | None = None) -> httpx.BaseTransport:
    return CurlTransport(  # or AsyncCurlTransport
      proxy=proxy,
      verify=False,  # and other custom options
    )

client = httpx.Client(  # or httpx.AsyncClient
    transport=transport_factory(),
    mounts={
        k: transport_factory(httpx.Proxy(url=v)) if v else None
        for k, v in get_environment_proxies().items()
    }
)

• httpx.Request content completely read in memory before sending, not sure if it's fixable with curl_cffi at all
• CurlTransport.cert argument should support in-memory data instead of filenames, pathlib.Path (instead of strings in httpx._types.CertTypes) is forced

• httpx.Timeout.pool is ignored, should be implemented in curl_cffi
• Simultaneous asynchronous requests requires to set CurlTransport.curl_options={CurlOpt.FRESH_CONNECT: True} lexiforest/curl_cffi#302 lexiforest/curl_cffi#319

• httpx.Timeout.write is ignored (libcurl limitation)

• CurlTransport.verify as ssl.SSLContext isn't supported (because OpenSSL is not used)

• CurlTransport.trust_env argument is ignored, libcurl is always using environment variables for configuration, which is disabled for proxies using CurlOpt.NOPROXY setting to make proxy argument have complete control on proxy usage, but may have effect in TLS configuration (but may not be used by curl-impersonate fork, idk) lexiforest/curl_cffi#345

• httpx.Response.request.headers isn't updated with default curl-impersonate headers, which can be unexpected on CurlTransport.default_headers=True lexiforest/curl_cffi#368

• CurlTransport.cert argument isn't compatible with (deprecated) httpx._types.CertTypes - impossible to pass password as third tuple element, pathlib.Path (instead of strings in httpx._types.CertTypes) is forced