Comment by vedaant00

Good question. PyScrappy's shared response cache is LRU-bounded with a configurable max size (cache_max_size, default 512), so it won't grow indefinitely even for a long-running MCP server hitting many distinct URLs. Oldest entries are evicted at the cap, expired ones are dropped on access, and reads promote to most-recently-used, so it stays bounded regardless of access pattern. Just shipped this in the latest release.