Computer science testbeds often require extensive experiment automation to be used efficiently. Jupyter notebooks can contain the scientific reasoning, experiment orchestration, and experiment results in an easy to read, easy to execute format. However some level of support is required from the testbeds to facilitate notebook use on their platforms. Additionally this format can be used for more than driving experiments, and different uses have different requirements in terms of support. In this paper we present an analysis of different notebook uses to be expected on computer science testbeds, as found through the feedback of users of the Grid'5000 testbed and a survey of other similar testbeds. Then we present the technical implementation of Jupyter on the Grid'5000 to support those uses.