Your approach is great, but won't work specifically with managed Kafka Connect. When you submit a JAR as a plugin, it doesn't get added to the classpath of kafka connect worker. It goes into a plugin dir. Each plugin has it's own classloader for isolation of the dependencies, etc. This is how Kafka Connect workers are using plugins.

So, this means that the MirrorSource Connector won't see the libraries and classes for a custom replication policy.

Alternatively, if you are just trying to achieve same topic name replication, there is an IdentityReplicationPolicy available in Kafka 3. To use it, define your MM2 connector to run in MSK Connect using version 3.7