When you browse GitHub or GitLab you can start a browser based integrated development environment with just one click. You can then start hacking with only a web browser and afterwards commit your results back to the project.
What needs to be done to do the same with OpenRefine in the context of datasets?
As a user you browse a collection of datasets and with one click you have a dataset loaded as OpenRefine project in your browser. With another click the updated data is sent back to the data provider.
As a "data provider" you can start an OpenRefine instance for just a single user/session and populate it with a given dataset. After working on this dataset there is some callback to update the dataset.