Random mutagenesis and deep mutational scanning (DMS) are widely used to optimize proteins by oversampling large libraries in microbial cells, selecting cells expressing favorable variants, and sequencing to identify enriched variants. However, these methods are slow and costly, require effort to establish selection methods for a given protein function, and do not yield data on lower-performing variants. Here, we describe Microbe-Independent Deep Assembly and Screening (MIDAS), a rapid, high-throughput method for optimizing protein function directly in mammalian cells. As a demonstration, we applied MIDAS to improve a newly designed neurotransmitter bioluminescent indicator (NeuBI) for acetylcholine (ACh). MIDAS systematically optimized interdomain linkers, identified mutational hotspots, and exhaustively scanned amino acid combinations, in each case relating specific sequences to protein performance. MIDAS-optimized variants exhibited improved performance in vivo, highlighting the potential of MIDAS for improving protein function in mammalian systems.