Save 60-80% tokens when AI reads code — MCP server for token-efficient code navigation with AST-aware structural reading
Digital-Threads / token-pilot
