PycURL Installation

NOTE: You need Python and libcurl installed on your system to use or build pycurl. Some RPM distributions of curl/libcurl do not include everything necessary to build pycurl, in which case you need to install the developer specific RPM which is usually called curl-dev.


Build and install pycurl with the following commands:

(if necessary, become root)
tar -zxvf pycurl-$VER.tar.gz
cd pycurl-$VER
python install

$VER should be substituted with the pycurl version number, e.g. 7.10.5.

Note that the installation script assumes that ‘curl-config’ can be located in your path setting. If curl-config is installed outside your path or you want to force installation to use a particular version of curl-config, use the ‘–curl-config’ command line option to specify the location of curl-config. Example:

python install --curl-config=/usr/local/bin/curl-config

If libcurl is linked dynamically with pycurl, you may have to alter the LD_LIBRARY_PATH environment variable accordingly. This normally applies only if there is more than one version of libcurl installed, e.g. one in /usr/lib and one in /usr/local/lib.


PycURL requires that the SSL library that it is built against is the same one libcurl, and therefore PycURL, uses at runtime. PycURL’s uses curl-config to attempt to figure out which SSL library libcurl was compiled against, however this does not always work. If PycURL is unable to determine the SSL library in use it will print a warning similar to the following:

src/pycurl.c:137:4: warning: #warning "libcurl was compiled with SSL support, but configure could not determine which " "library was used; thus no SSL crypto locking callbacks will be set, which may " "cause random crashes on SSL requests" [-Wcpp]

It will then fail at runtime as follows:

ImportError: pycurl: libcurl link-time ssl backend (openssl) is different from compile-time ssl backend (none/other)

To fix this, you need to tell what SSL backend is used:

python --with-[openssl|gnutls|nss] install

Note: as of PycURL 7.21.5, accepts --with-openssl option to indicate that libcurl is built against OpenSSL. --with-ssl is an alias for --with-openssl and continues to be accepted for backwards compatibility.

You can also ask to obtain SSL backend information from installed libcurl shared library, as follows:

python –

An unqualified would use the system libcurl, or you can specify a full path.

easy_install / pip

easy_install pycurl
pip install pycurl

If you need to specify an alternate curl-config, it can be done via an environment variable:

export PYCURL_CURL_CONFIG=/usr/local/bin/curl-config
easy_install pycurl

The same applies to the SSL backend, if you need to specify it (see the SSL note above):

export PYCURL_SSL_LIBRARY=[openssl|gnutls|nss]
easy_install pycurl

pip and cached pycurl package

If you have already installed pycurl and are trying to reinstall it via pip with different SSL options for example, pip may reinstall the package it has previously compiled instead of recompiling pycurl with newly specified options. More details are given in this Stack Overflow post.

To force pip to recompile pycurl, run:

# upgrade pip if necessary
pip install --upgrade pip

# remove current pycurl
pip remove pycurl


# recompile and install pycurl
pip install --compile pycurl


Official Packages

As of version 7.43.0, PycURL provides binary wheels for Windows. If you are using an official distribution of Python (i.e., one downloaded from, and you are using pip, you should be able to install PycURL by running:

pip install pycurl

If you are not using pip, EXE and MSI installers are available in the download area.

Both 32-bit and 64-bit builds of PycURL are available for Windows.

Using PycURL With Custom Python Builds

As of version 7.21.5, PycURL is linked statically against all of its dependencies except MSVCRT. This means that as long as your custom Python build uses the same version of MSVC as the corresponding official Python build as well as the same MSVCRT linking setting (/MD et. al.) you should be able to use an official PycURL package.

If your Python build uses different MSVCRT settings or a different MSVC version from the official Python builds, you will need to compile PycURL from source.

Currently official PycURL packages are built against the following Python versions:

  • 2.6.6
  • 2.7.10
  • 3.2.5
  • 3.3.5
  • 3.4.3
  • 3.5.0

If the C runtime library (MSVCRT.DLL) versions used by PycURL and Python do not match, you will receive a message like the following one when trying to import pycurl module:

ImportError: DLL load failed: The specified procedure could not be found.

To identify which MSVCRT version your Python uses use the application profiling feature of Dependency Walker and look for msvcrt.dll variants being loaded. You may find the entire thread starting here helpful.

Building From Source

Building PycURL from source is not for the faint of heart due to the multitude of possible dependencies. Additionally different dependencies have different settings for MSVCRT usage, and an application must have all of its parts agreeing on a single setting. If you decide to build PycURL from source you should familiarize yourself with the script - it is used to build the official binaries and tweaking it for your environment is likely to be less work than starting from scratch.

If you are compiling PycURL from source it is recommended to compile all of its dependencies from source as well. Using precompiled libraries may lead to multiple MSVCRT versions mixed in the resulting PycURL binary, which will not be good.

If PycURL is to be linked statically against its dependencies, OpenSSL must be patched to link to the DLL version of MSVCRT. There is a patch for this in winbuild directory of PycURL source.

For a minimum build you will just need libcurl source. Follow its Windows build instructions to build either a static or a DLL version of the library, then configure PycURL as follows to use it:

python --curl-dir=c:\dev\curl-7.33.0\builds\libcurl-vc-x86-release-dll-ipv6-sspi-spnego-winssl --use-libcurl-dll

Note that --curl-dir must point not to libcurl source but rather to headers and compiled libraries.

If libcurl and Python are not linked against the same exact C runtime (version number, static/dll, single-threaded/multi-threaded) you must use --avoid-stdio option (see below).

Additional Windows options:

  • --use-libcurl-dll: build against libcurl DLL, if not given PycURL will be built against libcurl statically.

  • --libcurl-lib-name=libcurl_imp.lib: specify a different name for libcurl import library. The default is libcurl.lib which is appropriate for static linking and is sometimes the correct choice for dynamic linking as well. The other possibility for dynamic linking is libcurl_imp.lib.

  • --with-openssl: use OpenSSL crypto locks when libcurl was built against OpenSSL.

  • --with-ssl: legacy alias for --with-openssl.

  • --avoid-stdio: on Windows, a process and each library it is using may be linked to its own version of the C runtime (MSVCRT). FILE pointers from one C runtime may not be passed to another C runtime. This option prevents direct passing of FILE pointers from Python to libcurl, thus permitting Python and libcurl to be linked against different C runtimes. This option may carry a performance penalty when Python file objects are given directly to PycURL in CURLOPT_READDATA, CURLOPT_WRITEDATA or CURLOPT_WRITEHEADER options. This option applies only on Python 2; on Python 3, file objects no longer expose C library FILE pointers and the C runtime issue does not exist. On Python 3, this option is recognized but does nothing. You can also give --avoid-stdio option in PYCURL_SETUP_OPTIONS environment variable as follows:

    PYCURL_SETUP_OPTIONS=--avoid-stdio pip install pycurl

A good target to use is bdist_wininst which produces an executable installer that you can run to install PycURL.

You may find the following mailing list posts helpful:

This script is used to build official PycURL Windows packages. You can use it to build a full complement of packages with your own options or modify it to build a single package you need.


  • Git for Windows.
  • Appropriate Python versions installed.
  • MS Visual C++ 9/2008 for Python <= 3.2, MS Visual C++ 10/2010 for Python 3.3 or 3.4, MS Visual C++ 14/2015 for Python 3.5. Express versions of Visual Studio work fine for this, although getting 64 bit compilers to wok in some Express versions involves jumping through several hoops.
  • NASM if building libcurl against OpenSSL.
  • ActivePerl if building libcurl against OpenSSL. The perl shipping with Git for Windows handles forward and backslashes in paths in a way that is incompatible with OpenSSL’s build scripts. assumes all programs are installed in their default locations, if this is not the case edit it as needed. itself can be run with any Python it supports - 2.6, 2.7 or 3.2 through 3.5.

Git Checkout

In order to build PycURL from a Git checkout, some files need to be generated. On Unix systems it is easiest to build PycURL with make:


To specify which curl or SSL backend to compile against, use the same environment variables as easy_install/pip, namely PYCURL_CURL_CONFIG and PYCURL_SSL_LIBRARY.

To generate generated files only you may run:

make gen

This might be handy if you are on Windows. Remember to run make gen whenever you change sources.

To generate documentation, run:

make docs

Generating documentation requires Sphinx to be installed.

A Note Regarding SSL Backends

libcurl’s functionality varies depending on which SSL backend it is compiled against. For example, users have reported problems with GnuTLS backend. As of this writing, generally speaking, OpenSSL backend has the most functionality as well as the best compatibility with other software.

If you experience SSL issues, especially if you are not using OpenSSL backend, you can try rebuilding libcurl and PycURL against another SSL backend.

SSL Certificate Bundle

libcurl, and PycURL, by default verify validity of HTTPS servers’ SSL certificates. Doing so requires a CA certificate bundle, which libcurl and most SSL libraries do not provide.

Here is a good resource on how to build your own certificate bundle. also has a prebuilt certificate bundle. To use the certificate bundle, use CAINFO or CAPPATH PycURL options.