I think that the line of argument against Putnamian anti-realism recently suggested by Paul Humphreys may well be extendable to areas beyond causal knowledge, but we need to think carefully about how Jim Fetzer's Peircean suggestions are playing a role here. Putnam's argument relies on the idea that our knowledge is based on (in) theories or other linguistic devices, and that theories are sets of sentences. Humphreys' suggestion via Fetzer is that if we can find a basis for knowledge that is non-propositional, we can avoid all of the notorious tangles that propositions create. Icons and indices are non-propositional, natural signs, so this looks like a promising route.
But what is crucial here, I think, is how we represent the information we get from the outside world. The intuition that causal processes are going on out in the world - and not in a language - is dead on, but what we need, it seems, is a way of showing how we cannot perform permutations on our representations of these processes. The representational relations that icons and indices enter into are between things in the outside world (unless we want to think of sensations and perceptions as icons and indices; more on this below perhaps). So the fact that icons and indices are natural signs (i.e., not constructed by us) is not going to be much help against the anti-realists unless our cognitions involving them are also non-propositional.