Optimize MCP query tools for large databases

[erikdarlingdata]

Summary

• MCP tools get_top_queries_by_cpu, get_top_procedures_by_cpu, and get_query_store_top timed out on large databases (600GB+, 100M+ rows) because they scanned every row, decompressed text on every row, and did client-side TOP N filtering
• Dashboard: New MCP-specific methods using multi-phase temp tables — aggregate numerics first, rank TOP N, hydrate text via OUTER APPLY for only the winners. DECOMPRESS now runs on ~N rows instead of the entire table
• Lite: Rewrote queries with CTE-based phasing — inner CTE aggregates numeric-only columns, outer query hydrates text via LEFT JOIN LATERAL for winners only
• Both: database_name filter and TOP N pushed into SQL instead of client-side .Where().Take()
• Shared UI methods are completely untouched

Performance

Server-side execution times on SQL2022 (all under 100ms):

Query	Avg Elapsed	Logical Reads
get_query_store_top	10ms	15,936
get_top_queries_by_cpu	87ms	112,162
get_top_queries_by_cpu (w/ db filter)	38ms	46,314
get_top_procedures_by_cpu	7ms	610

Test plan

• All 6 MCP tools tested (3 Lite + 3 Dashboard) — returning correct results with query text
• database_name filter tested — pushes into SQL correctly
• WAITFOR queries excluded via deferred filter on top N
• Dashboard UI tabs unaffected (shared methods untouched)
• Both projects build with 0 errors, 0 warnings

Fixes #826

🤖 Generated with Claude Code