The need of a surgical implantation of artificial material into the most delicate human organ is obviously the one big problem with the whole idea of a BCI that records brain activity by using implantable electrodes. Injuring the brain inevitably means creating a scar, and this can have an influence on normal brain function.
To minimize damage, the latest electrodes use glass-coated wires and other fancy materials, but in the end it all comes down to the size of what is to be implanted. This depends on the number of electrodes used or needed and the amount of "accessory material" which also has to be implanted, i.e. a wireless communication unit or, ideally, a whole chip with an integrated wireless interface that supports the neuroprosthesis.
The question about the number of electrodes is tricky: Most comments on the topic go for around 1000, but recent research suggests that far less might suffice, as long as real-time feedback is provided.

eeg / why is it a language / further nerve cell chatter
firing neurons / implantation / where the future lies
further chatter > implantation

Some background science here: Learn about eeg in medicine and why it can de used for BCI-control.