I'll be direct: once your vault passes about notes, Claude Code burns through tool calls and tokens searching for content. One team reported open-ended vault exploration ate through context fast enough to show on their bill.
Requests over tokens get charged at the normal input rate. Prompt shape matters more than prompt frequency.
The fix is targeting. Instead of "Tell me about my current projects" (scans everything), write "Read projects/index.md and summarize active items" (reads one file). Add index.md per folder to prevent blind scanning. For large vaults, Smart Connections MCP adds semantic search. Tool Search (January 2026) cuts MCP manifest overhead by %, compounding the savings.