Distributed software systems change dynamically due to the evolution of their environment and/or requirements, their internal designing policies, and/or their specification bugs which must be fixed. Hence, checking system changes must be run continuously. Such systems are usually composed of distributed software entities (called peers) interacting with each other through message exchanges, and this is to fulfil a common goal. The goal is often specified by a conversation protocol (CP), i.e. sequences of sent messages. If there exists a set of peers implementing CP, then CP is said to be realisable. In this paper, we propose a stepwise approach for checking whether an evolution, i.e. adding and/or removing messages and/or peers, can be applied to a CP that was realisable before updating it.We define a set of correct evolution patterns and we suggest an algebra of CP evolution. Our approach ensures that CP evolution preserves the realisability condition.