GDB knows the device running uCR as a ucr target. So, after starting GDB, attach it to uCR with a command such as
(uCR/i960) target ucr /dev/ise0
This will connect to the device /dev/ise0 as an attached processor device and will attempt to communicate with the GDB proxy there. When it connects to the proxy, it stops all the threads, selects one of the threads as the current thread, and returns to the user to wait for commands. GDB will select a thread that has faulted, if such a thread exists. If there are no faulted threads, then any one is selected.
When you are done with the target, either detach with the detach command, or quite GDB. In either case, GDB will communicate with the proxy to remote any breakpoints, will resume all the threads, and will detach from the proxy. The target then continues as normal.