Just to clarify some terms, which may actually muddy the water a bit. @mellowd's answer is a great short answer.
CLNS is the OSI connectionless network service definition: it specifies the interface that the service user (transport layer) uses to communicate with its peer transport layers. CLNP is the OSI protocol that provides that service, and is rarely used any more.
ISIS did not predate CLNS, but that's beside the point because ISIS can be used without CLNS. It is a part of the OSI network layer because it is used to help CLNP (or in our case IP, another network layer protocol) do its job. ISIS does not run over CLNS or CLNP; it runs directly over the link layers. (This is in contrast to OSPF, which runs over IP. ISIS is academically more correct, but OSPF is quicker to get up and running over a new kind of link. Whee.)
The term "clns mtu" is a bit misleading, though for reasonable reasons. Cisco uses "clns" to configure lots of things about the erstwhile OSI network layer, including some aspects of ISIS. So, in the bucket of stuff Cisco bundled together under "clns", they have "mtu" which means something like "Anything in the OSI network layer should assume this is the MSDU (maximum service data unit) of the underlying layer." That would dictate the maximum size of any network layer PDU, not counting the L2 overhead.
So, this "clns mtu" actually operates at the bottom of the OSI network layer, just between it and any link layer it runs over. But it's not an official term; it's just what Cisco calls it (and no doubt now lots of other implementations that use a similar configuration model.)
MSS is the maximum segment size for TCP. Since TCP is rarely run over CLNP (hopefully rfc-1006 and TUBA are dead), we'd never talk about the relationship between MSS and CLNS MTU. If we did, it'd be complicated because the CLNP header size is variable and depends on the address (NSAP) lengths, which can be up to 20 bytes, plus options, to a maximum header of 254 bytes. Let's not go there.
So, what is important? Well the ISIS LSP (link state PDU) size has to be the same thought any L1 or L2 area, so that an LSP issued by any router in the area can flood to all the others over any link. So, find the link with the smallest value for MTU minus L2 overhead, and use that value on all routers.
It's better to be conservative and it's not crucial to squeeze out every byte possible, in most cases. ISIS will segment its LSPs and be relatively efficient at disseminating the link state info. Feel free to leave room for dot1q tags or whatever, in case you'll be using tunnels between ISIS routers.