It's packaged up so it installs from the test toolshed directly via the admin interface of your test instance, but you'll need a good reliable network as there is a lot that gets downloaded and installed.
Like other toolshed tools, it would be possible but not advisable to download and unpack the archive at
http://testtoolshed.g2.bx.psu.edu/view/fubar/edger_test and install manually but your instance will need R 3.0.1 with bioconductor/edgeR/deseq/limma installed - if it's your first try, it will probably take some time and effort to read all the relevant docs to do that or the R/BioC mailing lists can provide assistance if you need it.
The edgeR tool requires a count matrix (reads per sample per contig) as input which the htseq tool
http://testtoolshed.g2.bx.psu.edu/view/fubar/htseq_bams_to_count_matrix will generate from any number of aligned bam/sam files and a GTF file with the appropriate gene model. Again, automatic installation from the toolshed is highly recommended, otherwise you'll need to manually install the tool and the dependencies.
Note that edgeR requires replicate samples in each comparison group so the absolute minimum is 4 bam files as input to the count matrix maker. p values from analyses without replicates are not interpretable in any biologically meaningful way and are meaningless in terms of the usual frequentist interpretation in my opinion.