I need to debug communication protocol over 50 digital channels cable and 2MS/s sample rate. What kind of tool would be best suited to do that task? What other aspects should I consider?
The reason you can't find a scope with enough channels is because you're looking for the wrong tool. Scopes are best for analog stuff, and ones with more than 4 full-featured channels are rare. Each channel on a scope has a relatively complicated, and thus expensive, high-bandwidth low-noise amplifier, so a scope with 50+ channels would be extremely expensive.
For a digital signal, you don't need the analog front-end, you just need a comparator for each channel. A scope-like instrument with lots of digital inputs is called a logic analyser. 32 channels is pretty common, and 50+ channel versions aren't too hard to find. As a bonus logic analysers are usually better than scopes at decoding the signals and displaying higher-level protocol-dependent information.
One thing to watch out for is that the logic analyser can't tell you as much about a signal which is wrong. It can't tell the difference between a signal which is sent wrong, and one which has been messed up by interference. So you still need the scope. Use the logic analyser to identify which signal(s) is/are not what they should be, then look with the scope to see if the problem lies in the digital logic or in the signal integrity.
What you probably want is a logic analyzer. There are a lot of "maker" quality ones available. Sigrok is a pretty common software package. If you look at the linked list, I don't think there are any with more than 34 channels.
If there are no tools for what you want to do, it is probably 1 of 2 things. 1. You are doing something silly because you don't know a better way of doing it. In that case it might be of value to tell people what you are doing so they can give you suggestions. 2. You are doing something so custom you are going to have to design your own test hardware.
It does look like there are logic analyzers with that many channels available from Agilent so 1 might not apply but I suspect it does.