curl_cffi

curl_cffi is a Python HTTP client that binds to a curl-impersonate fork through cffi, letting your requests mimic the TLS/JA3, HTTP/2, and even HTTP/3 fingerprints of real browsers. Unlike pure-Python clients such as requests or httpx, it presents a network fingerprint that anti-bot systems recognize as a genuine browser, so it often gets through where other clients are blocked.

It exposes a familiar requests-style API — plus an asyncio interface and WebSocket support — while handling the low-level libcurl binding for you. That makes it a popular choice for web scraping and automation against sites that fingerprint clients at the TLS layer.

What You Get

• A requests-style synchronous API (get, post, Session) backed by libcurl
• Browser impersonation of TLS/JA3, JA4, HTTP/2, and HTTP/3 fingerprints via an impersonate parameter
• An asyncio client plus WebSocket support for concurrent and streaming workloads
• Proxy support including SOCKS5, and a CLI for debugging fingerprints
• Prebuilt wheels bundling the native curl-impersonate library across platforms

Common Use Cases

• Web scraping sites that block clients by TLS/JA3 fingerprint
• Automating requests that must appear to originate from a real browser
• High-throughput async HTTP calls that need curl-level performance
• Replacing requests/httpx when anti-bot systems reject pure-Python clients

Under The Hood

Architecture - The curl_cffi package layers a Pythonic API over libcurl. curl.py holds the low-level cffi binding to the bundled curl-impersonate build, fingerprints.py encodes the browser TLS/JA3/HTTP2 presets, and the requests/ subpackage provides the high-level requests-style Session, get/post helpers, and response objects. aio.py and _asyncio_selector.py add the asyncio client, and a cli subpackage exposes a debugging command. Native sources live under ffi/ with libs.json describing the prebuilt binaries.

Tech Stack - Python 3.10+ with just two runtime dependencies — cffi>=2.0.0 for the C binding and certifi for CA bundles. The heavy lifting is done by the vendored curl-impersonate C library, shipped as platform wheels built with cibuildwheel. Dev tooling includes pytest, pytest-asyncio, trio, mypy, and ruff.

Code Quality - The project is mature and actively maintained: it ships py.typed, enforces mypy and ruff, and carries an extensive tests/ suite covering sync, async, and trio paths against a local uvicorn server. Pinned dev dependencies and CI wheel builds indicate careful release engineering.

API Design - The public surface intentionally mirrors requests so adoption is nearly free, with browser impersonation reduced to a single impersonate="chrome"-style keyword. Async and WebSocket clients follow the same conventions, and thorough ReadTheDocs documentation plus examples keep the learning curve low despite the native machinery underneath.