I suppose the answer depends on the data you are working with. For example, one should use specialized genotype callers when working with virus populations. However, in general there are a lot of genotype callers available: GATK, FreeBayes, VcfTools. To use one of these tools one has to provide reference sequence and alignment (in BAM format) or a list of mutations (usually in VCF format).

I would select one sequence from multiple alignment as a reference sequence (this could be the consensus sequence, if no reference is available). Then I would remap the sequences forming the alignment to the selected reference using bwa-sw or bowtie2. The resulting BAM file can be used to generate VCF file for genotyping.

I guess I am wondering why you would want to do this? You'll be able to easily identify a majority of genotypes by consensus, but rare variants may be hard to discern from sequencing errors.

You didn't provide us any information in regards to a research question or rationale for this question, so it's hard to provide some input on this basis.