can't believe i never spotted this thread before. well things have moved on a lot since I used MPTCP back in 2017, so I probably can't add much value.
But in summary I was working a on project to bond multiple connections together for vessels (superyachts etc...). 550ms Geo Satellite (only option at sea), 4G global SIM, or WiFi at Shore. I settled on MPTCP because the separate subflows allows the bonding to work seamlessly as the availability, bandwidth and latency changed on these connections over time and no manual intervention to switch links on/off. Also packet loss on one connections would not throttle the other connections.
I've since left that organisation and I believe they have worked on it and branded it much further now to include Starlink at sea also.
The way I was doing it was with OpenVPN tunnel and redsocks proxy to steer the connections through the tunnel to an AWS hosted "multiplexer" for internet breakout. This was needed at the time because most web servers didn't have MPTCP in the mainstream kernel. It was a complete DIY solution with a mini python web front end and Grafana dashboard to manage it.
Downside was AWS costs and increased latency steering traffic to AWS before local breakout to the internet, although this could be mitigated by picking the closest AWS node based on the vessels location.