curlpro

An HTTP client with a browser's network fingerprint: the TLS ClientHello, the HTTP/2 and HTTP/3 frames, header order and header case.

pip install curlpro

import curlpro

with curlpro.Session("chrome-151-windows") as s:
    r = s.get("https://example.com")
    print(r.status, r.text[:200])

Neither Go nor a compiler is needed: the native library and all 48 profiles are already inside the wheel, and the profiles load themselves.

Why another one

The existing clients keep their browser profiles in compiled code: a new Chrome comes out every four weeks, and each time that means editing C or Go, rebuilding and releasing. Here a profile is data, and it can be registered at runtime:

curlpro.register_profile({
    "name": "chrome-152-windows",
    "based_on": "chrome-151-windows",
    "headers": {"user_agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) ... Chrome/152.0.0.0 ..."},
})

Your own fingerprint, without a request

What a server would see is computed locally, from the same ClientHello bytes that would go on the wire. No network, no oracle:

with curlpro.Session("chrome-151-windows") as s:
    fp = s.fingerprint()
    print(fp.ja4)       # t13d1516h2_8daaf6152771_806a8c22fdea
    print(fp.akamai)    # 1:65536;2:0;4:6291456;6:262144|15663105|0|m,a,s,p

Checked against 48 captures: JA4 48/48, JA3N 48/48, Akamai 48/48.

audit() answers the second question — the one people actually lose days to. Not "does my fingerprint look right" but "does anything here disagree with anything else", because that is what gives a client away:

for finding in s.audit():
    print(finding)

Personas

Profile, proxy, device, headers and cookies — one identity, one file:

p = curlpro.Persona.new("chrome-151-windows", proxy="http://user:pass@host:8080")
p.save("accounts/user42.json")

p = curlpro.Persona.load("accounts/user42.json")
with p.session() as s:
    s.get("https://example.com/")
p.save()          # the cookies moved on; the identity did not

requests compatibility

import curlpro.requests as requests

r = requests.get("https://example.com/", timeout=10)

Existing code changes one import. It is a subset, and it says so: an argument the shim cannot honour is refused with a reason rather than ignored.

What is inside

A thin ctypes wrapper over a native library written in Go: the handshake is driven by uTLS, HTTP/2 by fhttp, QUIC by uquic.

The fingerprint is checked against tls.browserleaks.com: Chrome 151 gives t13d1516h2_8daaf6152771_806a8c22fdea — the same JA4 as the live browser.

Boundaries

The library covers the network layer. It does not forge the JS fingerprint (canvas, WebGL, navigator) — that is the browser's level, and the answer there is Playwright. Matching the network fingerprint is necessary but not sufficient: modern systems score JA4 together with JA4H, JA3S/JARM and behaviour.

Cleartext http:// and ws:// work too. There is no ClientHello and so no TLS fingerprint there, but the HTTP/1.1 half of the profile still applies — the header order and case — and that is all a plain-HTTP peer can see anyway. The point is not masking: a caller whose own service speaks plain HTTP, a solver or an internal API, should not need a second HTTP client beside this one.

HTTP/1.1, HTTP/2, HTTP/3 and WebSocket are supported, along with cookies, redirects, proxies (HTTP CONNECT and SOCKS5), multipart, streaming reads and uploads, and an asynchronous API.

# WebSocket: the handshake follows the profile's template, permessage-deflate works
with curlpro.Session() as s:
    with s.websocket("wss://echo.websocket.org/", max_message_size=1 << 20) as ws:
        ws.send("hello")         # str   -> a text frame
        ws.send(b"\x00\xff")     # bytes -> a binary one
        for message in ws:       # until the server closes: curlpro.WebSocketClosed;
            print(message)       # a silence timeout is CurlProError with .code == "timeout"

# A large file goes as a stream rather than through memory
with curlpro.Session() as s:
    s.post("https://example.com/upload", body_file="archive.zip")

# The connection is reused between requests, as a browser's is. keep_alive=False
# gives every request its own — needed when a balancer pins a client to one node.
with curlpro.Session(keep_alive=False) as s:
    s.get("https://example.com/")

The HTTP/3 fingerprint is checked against Chrome 144 on quic.browserleaks.com:

with curlpro.Session("chrome-151-windows", http3=True) as s:
    print(s.get("https://quic.browserleaks.com/fp").json()["h3_text"])
    # 1:65536;6:262144;7:100;51:1;GREASE|GREASE|984832|m,a,s,p

The QPACK dynamic table is supported by a decoder of our own: the profile advertises a capacity as Chrome does, and a server that uses it gets parsed.

JA4H can be left out of the build: it is the one component under a different licence (FoxIO License 1.1, patent-pending), and -tags nofoxio excludes it while leaving every other fingerprint intact. fingerprint().ja4h_available says which build is in use. The wheels here are built with it.

Install

pip install curlpro

Nothing to compile: the native library and all 48 profiles travel inside the wheel. Wheels are built for Linux (x86-64 and ARM64, glibc 2.28+), macOS 13+ (Intel and Apple Silicon) and Windows x64. The macOS 13 floor is not ours to choose: that is what Go 1.27 requires, and the native part is built with it.

Platforms outside that list — Alpine and other musl distributions, Windows on ARM, older glibc or macOS — have no wheel. There the source archive is built by hand, and Go and a C compiler are required: pip install alone would leave the package without its native part, and the failure would come at the first call rather than at install time.

pip download curlpro --no-binary :all: --no-deps
tar -xzf curlpro-*.tar.gz && cd curlpro-*/go
CGO_ENABLED=1 go build -buildmode=c-shared -o ../curlpro/lib/libcurlpro.so ./lib

The library is looked up through CURLPRO_LIBRARY, then in curlpro/lib/, then in dist/.

Full documentation and sources — github.com/int3re/curlpro.