Paparats – Local Semantic Code Search MCP

Built for AI coding assistants that need to understand your entire codebase, not just open files. Paparats indexes repositories locally using AST-aware chunking via tree-sitter, stores embeddings in Qdrant, and exposes semantic search plus symbol graph traversal through MCP tools. The server tracks real call/reference relationships across chunks so your agent can trace dependencies without grepping. It runs Ollama embeddings on your machine by default, keeps git blame metadata per chunk, and maintains an architectural memory store where the agent writes and reads ADRs as it learns your system. The operator console at /ui shows token savings, query patterns, and cross-project usage. Works with Claude, Cursor, Windsurf, or any MCP client that needs deep code context across multiple repos without sending anything to the cloud.