Filesift

Exposes local codebase search through four MCP tools: filesift_search for natural language queries like "authentication middleware", filesift_find_related for discovering connected files via imports and semantic similarity, filesift_index for creating searchable indexes, and filesift_status for checking index state. Uses hybrid BM25 and FAISS vector search with reciprocal rank fusion to balance keyword matching and semantic understanding. Runs a background daemon that keeps embeddings in memory for fast queries after initial cold start. Everything stays local. Indexes live in a .filesift directory per project. Reach for this when your agent wastes tokens exploring codebases or grep falls short on intent-based searches.