Abstract
LLM agents are now a primary interface to data repositories, but they locate files by issuing speculative regex queries through glob, grep, and file reads, creating a guess-and-check loop that is expensive in tool calls and tokens and ignores metadata embedded in scientific file formats. We present Acropolis, a semantic search system that augments a tiered filesystem with adaptive-depth content indexing and exposes it to agents through a typed natural-language interface. Acropolis provides a transparent file-summary operator in the I/O ingest path with content-addressable idempotency, a pluggable search-backend abstraction for lexical, vector, and graph engines, and an MCP server for compatible agents. On a 70-query CMIP6 retrieval benchmark, an agent using Acropolis matches the shell-tool baseline on retrieval accuracy while substantially reducing tool and backend calls per query and LLM token consumption.