Once you really have hundreds of files, it's time to consider a variant warehouse. Then you will be spending your time querying instead of manipulating VCF files.

One open source warehouse is TileDB-VCF (repo)

import tiledb
import tiledbvcf

vcf_uri = "my_vcf_dataset"
ds = tiledbvcf.Dataset(uri="mytiledbvcfstore", mode="w")
ds.create_dataset(enable_variant_stats=True, enable_allele_count=True) 
ds.ingest_samples(sample_uris = "input.vcf.gz")

df = ds.read(
   regions = ["chr17:63480412-63480412","chr17:63481680-63481680"]
   attrs   = ["sample_name", "contig", "pos_start", "pos_end", "alleles", "fmt_GT"],
)
df

We do have some scripts that can help with VEP annotation on TileDB arrays

More info here (blog) and here (docs)

A the manual states you can use the -r regions, or the -R, regions-file parameters:

• https://samtools.github.io/bcftools/bcftools.html

it would look like:

bcftools view -r chr17:63480412,chr17:63481680 input.vcf -o output.vcf

or

bcftools view -R regions.bed input.vcf -o output.vcf

now to operate on all files, use gnu parallel, assuming your filenames are in a file, it would look something like:

cat filenames.txt | parallel bcftools view -R regions.bed {} -o {}.small.vcf