Skip to main content

A minimal PyPI server for use with pip/easy_install.

Project description


pypiserver - minimal PyPI server for use with pip/easy_install

Latest Version in PyPI test status Dependencies up-to-date? Supported Python versions Project License








Kostis Anagnostopoulos <>,
Matthew Planchard <>,
Dmitrii Orlov <>,
Someone new? We are looking for new maintainers! <>

zlib/libpng + MIT


Chat with us on Zulip!

pypiserver is a minimal PyPI compatible server for pip or easy_install. It is based on bottle and serves packages from regular directories. Wheels, bdists, eggs and accompanying PGP-signatures can be uploaded either with pip, setuptools, twine, pypi-uploader, or simply copied with scp.

Quickstart: Installation and Usage

pypiserver works with Python 3.6+ and PyPy3.

Older Python versions may still work, but they are not tested.

For legacy Python versions, use pypiserver-1.x series. Note that these are not officially supported, and will not receive bugfixes or new features.

  1. Install pypiserver with this command:

    pip install pypiserver                # Or: pypiserver[passlib,cache]
    mkdir ~/packages                      # Copy packages into this directory.

    See also Alternative Installation methods.

  2. Copy some packages into your ~/packages folder and then get your pypiserver up and running:

    pypi-server -p 8080 ~/packages &      # Will listen to all IPs.
  3. From the client computer, type this:

    # Download and install hosted packages.
    pip install --extra-index-url http://localhost:8080/simple/ ...
    # or
    pip install --extra-index-url http://localhost:8080 ...
    # Search hosted packages.
    pip search --index http://localhost:8080 ...
    # Note that pip search does not currently work with the /simple/ endpoint.

    See also Client-side configurations for avoiding tedious typing.

  4. Enter pypi-server -h in the cmd-line to print a detailed usage message:

    pypi-server [OPTIONS] [PACKAGES_DIRECTORY...]
      start PyPI compatible package server serving packages from
      command line, it uses the default ~/packages. pypiserver scans this
      directory recursively for packages. It skips packages and
      directories starting with a dot. Multiple package directories can be
    pypi-server understands the following options:
      -p, --port PORT
        Listen on port PORT (default: 8080).
      -i, --interface INTERFACE
        Listen on interface INTERFACE (default:, any interface).
      -a, --authenticate (update|download|list), ...
        Comma-separated list of (case-insensitive) actions to authenticate.
        Requires to have set the password (-P option).
        To password-protect package downloads (in addition to uploads) while
        leaving listings public, use:
          -P foo/htpasswd.txt -a update,download
        To allow unauthorized access, use:
          -P . -a .
        Note that when uploads are not protected, the `register` command
        is not necessary, but `~/.pypirc` still need username and password fields,
        even if bogus.
        By default, only 'update' is password-protected.
      -P, --passwords PASSWORD_FILE
        Use apache htpasswd file PASSWORD_FILE to set usernames & passwords when
        authenticating certain actions (see -a option).
        To allow unauthorized access, use:
          -P . -a .
        Disable redirect to real PyPI index for packages not found in the
        local index.
      --fallback-url FALLBACK_URL
        For packages not found in the local index, this URL will be used to
        redirect to (default:
      --server METHOD
        Use METHOD to run the server. Valid values include paste,
        cherrypy, twisted, gunicorn, gevent, wsgiref, auto. The
        default is to use "auto" which chooses one of paste, cherrypy,
        twisted or wsgiref.
      -r, --root PACKAGES_DIRECTORY
        [deprecated] Serve packages from PACKAGES_DIRECTORY.
      -o, --overwrite
        Allow overwriting existing package files.
      --hash-algo ALGO
        Any `hashlib` available algo used as fragments on package links.
        Set one of (0, no, off, false) to disabled it (default: md5).
      --welcome HTML_FILE
        Uses the ASCII contents of HTML_FILE as welcome message response.
        Enable verbose logging; repeat for more verbosity.
      --log-conf <FILE>
        Read logging configuration from FILE.
        By default, configuration is read from `log.conf` if found in server's dir.
      --log-file <FILE>
        Write logging info into this FILE.
      --log-frmt <FILE>
        The logging format-string (see `logging.LogRecord` class from standard python library).
        [Default: %(asctime)s|%(name)s|%(levelname)s|%(thread)d|%(message)s]
      --log-req-frmt FORMAT
        A format-string selecting Http-Request properties to log; set to '%s' to see them all.
        [Default: %(bottle.request)s]
      --log-res-frmt FORMAT
        A format-string selecting Http-Response properties to log; set to  '%s' to see them all.
        [Default: %(status)s]
      --log-err-frmt FORMAT
        A format-string selecting Http-Error properties to log; set to  '%s' to see them all.
        [Default: %(body)s: %(exception)s \n%(traceback)s]
      --cache-control AGE
        Add "Cache-Control: max-age=AGE, public" header to package downloads.
        Pip 6+ needs this for caching.
    pypi-server -h, --help
      Show this help message.
    pypi-server --version
      Show pypi-server's version.
    pypi-server -U [OPTIONS] [PACKAGES_DIRECTORY...]
      Update packages in PACKAGES_DIRECTORY. This command searches for updates and shows a pip command line which
      updates the package.
    The following additional options can be specified with -U:
        Execute the pip commands instead of only showing them.
        Download package updates to this directory. The default is to use
        the directory which contains the latest version of the package to
        be updated.
        Allow updating to unstable version (alpha, beta, rc, dev versions).
    Visit for more information.

Client-Side Configurations

Always specifying the the pypi url on the command line is a bit cumbersome. Since pypiserver redirects pip/easy_install to the index if it doesn’t have a requested package, it is a good idea to configure them to always use your local pypi index.

Configuring pip

For pip command this can be done by setting the environment variable PIP_EXTRA_INDEX_URL in your .bashr/.profile/.zshrc:

export PIP_EXTRA_INDEX_URL=http://localhost:8080/simple/

or by adding the following lines to ~/.pip/pip.conf:

extra-index-url = http://localhost:8080/simple/

Configuring easy_install

For easy_install command you may set the following configuration in ~/.pydistutils.cfg:

index_url = http://localhost:8080/simple/

Uploading Packages Remotely

Instead of copying packages directly to the server’s folder (i.e. with scp), you may use python tools for the task, e.g. python upload. In that case, pypiserver is responsible for authenticating the upload-requests.

Apache-Like Authentication (htpasswd)

  1. First make sure you have the passlib module installed (note that passlib>=1.6 is required), which is needed for parsing the Apache htpasswd file specified by the -P, --passwords option (see next steps):

    pip install passlib
  2. Create the Apache htpasswd file with at least one user/password pair with this command (you’ll be prompted for a password):

    htpasswd -sc htpasswd.txt <some_username>
  3. You need to restart the server with the -P option only once (but user/password pairs can later be added or updated on the fly):

    ./pypi-server -p 8080 -P htpasswd.txt ~/packages &

Upload with setuptools

  1. On client-side, edit or create a ~/.pypirc file with a similar content:

    index-servers =
    repository: http://localhost:8080
    username: <some_username>
    password: <some_passwd>
  2. Then from within the directory of the python-project you wish to upload, issue this command:

    python sdist upload -r local

Upload with twine

To avoid storing you passwords on disk, in clear text, you may either:

  • use the register setuptools’s command with the -r option, like that:

    python sdist register -r local upload -r local
  • use twine library, which breaks the procedure in two steps. In addition, it supports signing your files with PGP-Signatures and uploading the generated .asc files to pypiserver:

    twine upload -r local --sign -identity user_name ./

Using the Docker Image

Starting with version 1.2.5, official Docker images will be built for each push to master, each dev, alpha, or beta release, and each final release. The most recent full release will always be available under the tag latest, and the current master branch will always be available under the tag unstable.

You can always check to see what tags are currently available at our Docker Repo.

To run the most recent release of pypiserver with Docker, simply:

docker run pypiserver/pypiserver:latest

This starts pypiserver serving packages from the /data/packages directory inside the container, listening on the container port 8080.

The container takes all the same arguments as the normal pypi-server executable, with the exception of the internal container port (-p), which will always be 8080.

Of course, just running a container isn’t that interesting. To map port 80 on the host to port 8080 on the container:

docker run -p 80:8080 pypiserver/pypiserver:latest

You can now access your pypiserver at localhost:80 in a web browser.

To serve packages from a directory on the host, e.g. ~/packages:

docker run -p 80:8080 -v ~/packages:/data/packages pypiserver/pypiserver:latest

To authenticate against a local .htpasswd file:

docker run -p 80:8080 -v ~/.htpasswd:/data/.htpasswd pypiserver/pypiserver:latest -P .htpasswd packages

You can also specify pypiserver to run as a Docker service using a composefile. An example composefile is provided.

Alternative Installation Methods

When trying the methods below, first use the following command to check whether previous versions of pypiserver already exist, and (optionally) uninstall them:

# VERSION-CHECK: Fails if not installed.
pypi-server --version

# UNINSTALL: Invoke again untill it fails.
pip uninstall pypiserver

Installing the Very Latest Version

In case the latest version in pypi is a pre-release, you have to use pip’s –pre option. And to update an existing installation combine it with –ignore-installed:

pip install pypiserver --pre -I

You can even install the latest pypiserver directly from github with the following command, assuming you have git installed on your PATH:

pip install git+git://

Running on Heroku/Dotcloud contains instructions on how to run pypiserver on one of the supported cloud service providers.


Managing the Package Directory

The pypi-server command has the -U option that searches for updates of available packages. It scans the package directory for available packages and searches on for updates. Without further options pypi-server -U will just print a list of commands which must be run in order to get the latest version of each package. Output looks like:

$ ./pypi-server -U
checking 106 packages for newer version


no releases found on pypi for PyXML, Pymacs, mercurial, setuptools

# update raven from 1.4.3 to 1.4.4
pip -q install --no-deps  --extra-index-url -d /home/ralf/packages/mirror raven==1.4.4

# update greenlet from 0.3.3 to 0.3.4
pip -q install --no-deps  --extra-index-url -d /home/ralf/packages/mirror greenlet==0.3.4

It first prints for each package a single character after checking the available versions on pypi. A dot(.) means the package is up-to-date, 'u' means the package can be updated and 'e' means the list of releases on pypi is empty. After that it shows a pip command line which can be used to update a one package. Either copy and paste that or run pypi-server -Ux in order to really execute those commands. You need to have pip installed for that to work however.

Specifying an additional -u option will also allow alpha, beta and release candidates to be downloaded. Without this option these releases won’t be considered.

Serving Thousands of Packages

By default, pypiserver scans the entire packages directory each time an incoming HTTP request occurs. This isn’t a problem for a small number of packages, but causes noticeable slow-downs when serving thousands of packages.

If you run into this problem, significant speedups can be gained by enabling pypiserver’s directory caching functionality. The only requirement is to install the watchdog package, or it can be installed during pypiserver installation, by specifying the cache extras option:

pip install pypiserver[cache]

Additional speedups can be obtained by using your webserver’s builtin caching functionality. For example, if you are using nginx as a reverse-proxy as described below in Behind a reverse proxy, you can easily enable caching. For example, to allow nginx to cache up to 10 gigabytes of data for up to 1 hour:

proxy_cache_path /data/nginx/cache

server {
    # ...
    location / {
        proxy_cache pypiserver_cache;
        proxy_pass http://localhost:8080;

Using webserver caching is especially helpful if you have high request volume. Using nginx caching, a real-world pypiserver installation was able to easily support over 1000 package downloads/min at peak load.

Managing Automated Startup

There are a variety of options for handling the automated starting of pypiserver upon system startup. Two of the most common are systemd and supervisor for linux systems. For windows creating services with scripts isn’t an easy task without a third party tool such as NSSM.

Running As a systemd Service

systemd is installed by default on most modern Linux systems and as such, it is an excellent option for managing the pypiserver process. An example config file for systemd can be seen below:

Description=A minimal PyPI server for use with pip/easy_install.

# systemd requires absolute path here too.

ExecStart=/usr/local/bin/pypi-server -p 8080 -a update,download --log-file /var/log/pypiserver.log -P /etc/nginx/.htpasswd /var/www/pypi
ExecStop=/bin/kill -TERM $MAINPID
ExecReload=/bin/kill -HUP $MAINPID




Adjusting the paths and adding this file as pypiserver.service into your systemd/system directory will allow management of the pypiserver process with systemctl, e.g. systemctl start pypiserver.

More useful information about systemd can be found at

Launching through supervisor

supervisor has the benefit of being a pure python package and as such, it provides excellent cross-platform support for process management. An example configuration file for supervisor is given below:

command=/home/pypi/pypi-venv/bin/pypi-server -p 7001 -P /home/pypi/.htpasswd /home/pypi/packages

From there, the process can be managed via supervisord using supervisorctl.

Running As a service with NSSM (Windows)

Download NSSM from unzip to a desired location such as Program Files. Decide whether you are going to use win32 or win64, and add that exe to environment PATH.

Create a start_pypiserver.bat:

pypi-server -p 8080 C:\Path\To\Packages &

Test the batch file by running it first before creating the service. Make sure you can access the server remotely, and install packages. If you can, proceed, if not troubleshoot until you can. This will ensure you know the server works, before adding NSSM into the mix.

From the command prompt:

nssm install pypiserver

This command will launch a NSSM gui application:

Path: C:\Path\To\start_pypiserver.bat
Startup directory: Auto generates when selecting path
Service name: pypiserver

There are more tabs, but that is the basic setup. If the service needs to be running with a certain login credentials, make sure you enter those credentials in the logon tab.

Start the service:

nssm start pypiserver

Other useful commands:

nssm --help
nssm stop <servicename>
nssm restart <servicename>
nssm status <servicename>

For detailed information please visit

Using a Different WSGI Server

  • The bottle web-server which supports many WSGI-servers, among others, paste, cherrypy, twisted and wsgiref (part of Python); you select them using the --server flag.

  • You may view all supported WSGI servers using the following interactive code:

    >>> from pypiserver import bottle
    >>> list(bottle.server_names.keys())
    ['cgi', 'gunicorn', 'cherrypy', 'eventlet', 'tornado', 'geventSocketIO',
    'rocket', 'diesel', 'twisted', 'wsgiref', 'fapws3', 'bjoern', 'gevent',
    'meinheld', 'auto', 'aiohttp', 'flup', 'gae', 'paste', 'waitress']
  • If none of the above servers matches your needs, invoke just the pypiserver:app() method which returns the internal WSGI-app WITHOUT starting-up a server - you may then send it to any WSGI server you like. Read also the Utilizing the API section.

  • Some examples are given below - you may find more details in bottle site.

Apache (mod_wsgi)

To use your Apache2 with pypiserver, prefer to utilize mod_wsgi as explained in bottle’s documentation.

  1. Adapt and place the following Apache configuration either into top-level scope, or inside some <VirtualHost> (contributed by Thomas Waldmann):

    WSGIScriptAlias   /     /yoursite/wsgi/
    WSGIDaemonProcess       pypisrv user=pypisrv group=pypisrv umask=0007 \
                            processes=1 threads=5 maximum-requests=500 \
                            display-name=wsgi-pypisrv inactivity-timeout=300
    WSGIProcessGroup        pypisrv
    WSGIPassAuthorization On    # Required for authentication (
    <Directory /yoursite/wsgi >
        Require all granted

    or if using older Apache < 2.4, substitute the last part with this:

    <Directory /yoursite/wsgi >
        Order deny,allow
        Allow from all
  2. Then create the /yoursite/cfg/pypiserver.wsgi file and make sure that the user and group of the WSGIDaemonProcess directive (pypisrv:pypisrv in the example) have the read permission on it:

    import pypiserver
    conf = pypiserver.default_config(
        root =          "/yoursite/packages",
        password_file = "/yoursite/htpasswd", )
    application =**conf)

The following command uses gunicorn to start pypiserver:

gunicorn -w4 'pypiserver:app(root="/home/ralf/packages")'

or when using multiple roots:

gunicorn -w4 'pypiserver:app(root=["/home/ralf/packages", "/home/ralf/experimental"])'

paste allows to run multiple WSGI applications under different URL paths. Therefore it is possible to serve different set of packages on different paths.

The following example paste.ini could be used to serve stable and unstable packages on different paths:

use = egg:Paste#urlmap
/unstable/ = unstable
/ = stable

use = egg:pypiserver#main
root = ~/stable-packages

use = egg:pypiserver#main
root = ~/stable-packages

use = egg:gunicorn#main
host =
port = 9000
workers = 5
accesslog = -

Behind a Reverse Proxy

You can run pypiserver behind a reverse proxy as well.


Extend your nginx configuration:

upstream pypi {
  server     fail_timeout=0;

server {

  location / {
    proxy_set_header  Host $host:$server_port;
    proxy_set_header  X-Forwarded-Proto $scheme;
    proxy_set_header  X-Real-IP $remote_addr;
    proxy_pass        http://pypi;

As of pypiserver 1.3, you may also use the X-Forwarded-Host header in your reverse proxy config to enable changing the base URL. For example if you want to host pypiserver under a particular path on your server:

upstream pypi {
  server              localhost:8000;

server {
  location /pypi/ {
      proxy_set_header  X-Forwarded-Host $host:$server_port/pypi;
      proxy_set_header  X-Forwarded-Proto $scheme;
      proxy_set_header  X-Forwarded-For $proxy_add_x_forwarded_for;
      proxy_set_header  X-Real-IP $remote_addr;
      proxy_pass        http://pypi;
Supporting HTTPS

Using a reverse proxy is the preferred way of getting pypiserver behind HTTPS. For example, to put pypiserver behind HTTPS on port 443, with automatic HTTP redirection, using nginx:

upstream pypi {
  server               localhost:8000;

server {
  listen              80 default_server;
  server_name         _;
  return              301 https://$host$request_uri;

server {
  listen              443 ssl;

  ssl_certificate     /etc/;
  ssl_certificate_key /etc/;
  ssl_protocols       TLSv1 TLSv1.1 TLSv1.2;
  ssl_ciphers         HIGH:!aNULL:!MD5;

  location / {
    proxy_set_header  Host $host:$server_port;
    proxy_set_header  X-Forwarded-Proto $scheme;
    proxy_set_header  X-Real-IP $remote_addr;
    proxy_pass        http://pypi;

Please see nginx’s HTTPS docs for more details.

Getting and keeping your certificates up-to-date can be simplified using, for example, using certbot and letsencrypt.


It is also possible to use Traefik to put pypiserver behind HTTPS on port 443, with automatic HTTP redirection using Docker Compose. Please see the provided docker-compose.yml example for more information.

Utilizing the API

In order to enable ad-hoc authentication-providers or to use WSGI-servers not supported by bottle out-of-the-box, you needed to launch pypiserver via its API.

  • The main entry-point for configuring pypiserver is the pypiserver:app() function. This function returns the internal WSGI-app that you my then send to any WSGI-server you like.

  • To get all pypiserver:app() keywords and their explanations, read the function pypiserver:default_config().

  • Finally, to fire-up a WSGI-server with the configured app, invoke the bottle:run(app, host, port, server) function. Note that pypiserver ships with it is own copy of bottle; to use it, import it like that: from pypiserver import bottle

Using Ad-Hoc Authentication Providers

The auther keyword of pypiserver:app() function maybe set only using the API. This can be any callable that returns a boolean when passed the username and the password for a given request.

For example, to authenticate users based on the /etc/passwd file under Unix, you may delegate such decisions to the python-pam library by following these steps:

  1. Ensure python-pam module is installed:

    pip install python-pam
  2. Create a python-script along these lines:

    $ cat >
    import pypiserver
    from pypiserver import bottle
    import pam
    app ='./packages', auther=pam.authenticate), host='', port=80, server='auto')
    [Ctrl+ D]
  3. Invoke the python-script to start-up pypiserver:

    $ python

Use with MicroPython

The MicroPython interpreter for embedded devices can install packages with the module The module uses a specialized json-endpoint to retrieve package information. This endpoint is supported by pypiserver.

It can be tested with the UNIX port of micropython:

cd micropython
ports/unix/micropython -m tools.upip install -i http://my-server:8080 -p /tmp/mymodules micropython-foobar

Installing packages from the REPL of an embedded device works in this way:

import network
import upip

sta_if = network.WLAN(network.STA_IF)
sta_if.connect('<your ESSID>', '<your password>')
upip.index_urls = ["http://my-server:8080"]

Further information on micropython-packaging can be found here:


To create a copy of the repository, use:

git clone
cd pypiserver

To receive any later changes, in the above folder use:

git pull

Known Limitations

pypiserver does not implement the full API as seen on PyPI. It implements just enough to make easy_install, pip install, and search work.

The following limitations are known:

  • Command pypi -U that compares uploaded packages with pypi to see if they are outdated, does not respect a http-proxy environment variable (see #19).

  • It accepts documentation uploads but does not save them to disk (see #47 for a discussion)

  • It does not handle misspelled packages as pypi-repo does, therefore it is suggested to use it with --extra-index-url instead of --index-url (see #38).

Please use Github’s bugtracker for other bugs you find.

Similar Projects

There are lots of other projects, which allow you to run your own PyPI server. If pypiserver doesn’t work for you, the following are among the most popular alternatives:

Unmaintained or archived

These projects were once alternatives to pypiserver but are now either unmaintained or archived.

  • pip2pi a simple cmd-line tool that builds a PyPI-compatible local folder from pip requirements

  • flask-pypi-proxy A proxy for PyPI that also enables also uploading custom packages.


pypiserver contains a copy of bottle which is available under the MIT license, and the remaining part is distributed under the zlib/libpng license. See the LICENSE.txt file.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution (196.9 kB view hashes)

Uploaded Source

Built Distribution

pypiserver-1.5.0-py2.py3-none-any.whl (90.4 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page