Out of curiousity I just had a look at this document: http://puredata.info/community/projects/convention04/lectures/tk-grill/
This seems to be exactly what I need (with or without Python wrapper). Why is dyn~ better?
It's not that dyn~ is better, it's just different - it provides an easy way for in-patcher scripting, but for that it internally uses the current m_pd.h API. The dyn API is far more general, it is a proposal for a C interface to replace this current PD API, but at the moment it's not integrated into the PD kernel, therefore needing much more resources than necessary. This is ok for smaller applications but not for larger ones, as a replacement for a patcher system would be. However, you are welcome to use it and i'm happy about any feedback.
best greetings, Thomas