Building a Search & Fetch Layer for LLM Agents: Caching, Clean Extraction and Cost Control in Python