Splash and Jupyter

Splash provides a custom Jupyter (previously known as IPython) kernel for Lua. Together with Jupyter notebook frontend it forms an interactive web-based development environment for Splash Scripts with syntax highlighting, smart code completion, context-aware help, inline images support and a real live WebKit browser window with Web Inspector enabled, controllable from a notebook.


To install Splash-Jupyter using Docker, run:

$ docker pull scrapinghub/splash-jupyter

Then start the container:

$ docker run -p 8888:8888 -it scrapinghub/splash-jupyter


Without -it flags you won’t be able to stop the container using Ctrl-C.

If you’re on Linux, Jupyter server with Splash kernel enabled will be available at

If you use boot2docker, run $ boot2docker ip to get the ip address, the visit http://<ip-returned-by-boot2docker>:8888. If you use docker-machine, run $ docker-machine ip <your machine> to get the ip.

By default, notebooks are stored in a Docker container; they are destroyed when you restart an image. To persist notebooks you can mount a local folder to /notebooks. For example, let’s use current folder to store the notebooks:

$ docker run -v `/bin/pwd`/notebooks:/notebooks -p 8888:8888 -it splash-jupyter

Live Webkit window with web inspector is not available when Splash-Jupyter is executed from Docker. You can still use e.g. splash:png command to inspect what’s going on.

Currently to enable live Webkit window you must install Splash in a “manual way” - see Ubuntu 12.04 (manual way).

  1. Install IPython/Jupyter with notebook feature. Splash kernel requires IPython 4.x:

    $ pip install 'ipython[notebook] >= 4.0.0, < 5.0'
  2. Let IPython know about Splash kernel by running the following command:

    $ python -m splash.kernel install

To run IPython with Splash notebook, first start IPython notebook and then create a new Splash notebook using “New” button.

From Notebook to HTTP API

After you finished developing the script using a Jupyter Notebook, you may want to convert it to a form suitable for submitting to Splash HTTP API (see execute).

To do that, copy-paste (or download using “File -> Download as -> .lua”) all relevant code, then put it inside function main(splash):

function main(splash)
    -- Script code goes here,
    -- including all helper functions.
    return {...}  -- return the result

To make the script more generic you can use splash.args instead of hardcoded constants (e.g. for page urls). Also, consider submitting several requests with different arguments instead of running a loop in a script if you need to visit and process several pages - it is an easy way to parallelize the work.

There are some gotchas:

  1. When you run a notebook cell and then run another notebook cell there is a delay between runs; the effect is similar to inserting splash:wait calls at the beginning of each cell.
  2. Regardless of sandbox settings, scripts in Jupyter notebook are not sandboxed. Usually it is not a problem, but some functions may be unavailable in HTTP API if sandbox is enabled.