snowflake-mcp v22.0.6 → v22.0.9 (this session, 2026-05-08): 11 fixes/tools added across 3 batches. H
What
snowflake-mcp v22.0.6 → v22.0.9 (this session, 2026-05-08): 11 fixes/tools added across 3 batches. Highlights: (1) execute_query multi-statement support with paren/quote/dollar-aware scanner; (2) query_semantic_view fence-strip + IFF-form preservation + improved prompt template; (3) invoke_cortex_agent_v2 messages-array body fix + structured response (final_text, tool_used, sql_generated, tool_errors, suggested_queries — replaces 50KB SSE blob); (4) describe_semantic_view_v2(summary_only=True) and describe_cortex_agent(summary_only=True) compact modes; (5) NEW: list_semantic_view_metrics, list_semantic_view_dimensions, list_semantic_view_logical_tables, get_fivetran_sync_status, get_agent_full_ddl, validate_semantic_view_for_cortex. snowflake-report-mart skill bumped v1.0 → v1.1.0 with Hard Rules 6/7/8 + § Cortex Analyst formatting rules. 10 wrapper views deployed in RNWL_PROD.RNWL_SALES.*_C. Production: RNWL_SALES_DASHBOARDS bound to RUNWAL_BOT, agent now answers FY26 net sales by vertical correctly (₹2,357 Cr matches Tableau ₹2,242 within 5%).
Why
Routing decisions for Snowflake/Cortex Analyst work should leverage the new validator + structured response + introspection helpers. The 5-deployment-cycle Cortex 392700 debug pattern is now preventable at three layers (skill rules / validator tool / wrapper-view convention).
Action Required
Update routing knowledge: any future “build Cortex Analyst semantic view” task should call validate_semantic_view_for_cortex(ddl) before CREATE OR REPLACE; any agent-routing question on multi-tool agents should reference the verified-query-anchored discovery in session-learnings.md 2026-05-08.