Vector DB MCP

This server pairs DuckDB with assembly-accelerated vector operations for code search and symbol lookup. The underlying project uses FASM to write x86_64 dot product routines that get loaded as native libraries, falling back to pure Python if the compiled artifacts aren't available. The macOS port builds Mach-O dylibs instead of ELF shared objects, though you'll need Rosetta and an x86_64 Python runtime on Apple Silicon. Reach for this when you want fast semantic code search without spinning up a separate vector database process. The assembly acceleration targets the hot path in similarity calculations, which matters when you're indexing large codebases or running frequent symbol lookups against embedded vectors.