Can I use CountVectorizer in scikit-learn to count frequency of documents that were not used to extract the tokens?

You’re right that vocabulary is what you want. It works like this:

>>> cv = sklearn.feature_extraction.text.CountVectorizer(vocabulary=['hot', 'cold', 'old'])
>>> cv.fit_transform(['pease porridge hot', 'pease porridge cold', 'pease porridge in the pot', 'nine days old']).toarray()
array([[1, 0, 0],
       [0, 1, 0],
       [0, 0, 0],
       [0, 0, 1]], dtype=int64)

So you pass it a dict with your desired features as the keys.

If you used CountVectorizer on one set of documents and then you want to use the set of features from those documents for a new set, use the vocabulary_ attribute of your original CountVectorizer and pass it to the new one. So in your example, you could do

newVec = CountVectorizer(vocabulary=vec.vocabulary_)

to create a new tokenizer using the vocabulary from your first one.