I Measured What My Curated MCP Tool Output Is Actually Saving
python
dev.to
Every tool in my Developer Presence MCP server returns a hand-picked dict, never the raw API response. I did that from day one because it felt obviously correct — why hand a model twenty fields when it needs six. I never actually measured what it was saving. I finally ran the numbers this morning, on real data from my own account, and the size of the gap surprised me enough that I want to show the actual breakdown instead of just asserting "less is more." the tool in question list_ar