Codeseeker

This is a semantic code search engine that fuses BM25 keyword matching with vector embeddings and adds a knowledge graph on top. It exposes three MCP tools: search (with hybrid, vector-only, and full-text modes), analyze (for dependency traversal, coding standards detection, duplicate code, and dead code), and index. The search layer uses reciprocal rank fusion and includes RAPTOR directory summaries that automatically scope queries to relevant subdirectories. Graph expansion follows import, call, and extends edges to surface structurally related files even when they don't match the query text. Works with Claude Desktop, VS Code Copilot, and Cursor. Best for projects where you need AI assistants to understand architectural relationships, not just grep for strings.