UCSC Goldenpath offers 1kb, 2kb, and 5kb upstream sequences with annotated 5'UTR for various assemblies.
For hg38, for example, via http://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/:
upstream1000.fa.gz - Sequences 1000 bases upstream of annotated
transcription starts of RefSeq genes with annotated 5' UTRs.
This file is updated regularly. It might be slightly out of sync with
the RefSeq data shown on the browser, as is it updated daily for most assemblies.
upstream2000.fa.gz - Same as upstream1000, but 2000 bases.
upstream5000.fa.gz - Same as upstream1000, but 5000 bases.
For example, to download and expand the
upstream1000.fa.gz file for hg38:
$ wget -qO- "http://hgdownload.soe.ucsc.edu/goldenPath/hg38/bigZips/upstream1000.fa.gz" | gunzip -c > upstream1000.fa
Once you have selected sequences for your assembly, and you have a transcription factor PWM database in hand (Transfac, Jaspar, etc.) you could use FIMO to search for putative binding sites within the sequences.
On the Bioinformatics SE, I posted a walkthrough the commands to use to run FIMO with a JASPAR TF database and sequences-of-interest, at a typical threshold of sensitivity:
Another toolkit you might see mentioned is HOMER, but this is for de novo motif discovery, i.e., you are looking for new or unpublished motifs.
One difference between HOMER and FIMO is that FIMO would be used for discovery of published or known motifs, for which there are existing, experimentally validated PWM databases. The functionality of HOMER would perhaps be closer to the MEME tool, which is part of the larger toolkit that FIMO is in. Like HOMER, MEME would be used for finding novel motifs.