Second Brain

Surgical-minimum patch path for the 4 MCP gaps + isolated patches folder + Phase 0/1 prep deliverables built on top — chosen and shipped this session. snowflake-mcp v22.0.6 ships Gap 1 (drop CORTEX fr

2 min read

Surgical-minimum patch path for the 4 MCP gaps + isolated patches folder + Phase 0/1 prep deliverables built on top — chosen and shipped this session. snowflake-mcp v22.0.6 ships Gap 1 (drop CORTEX fr

Decision

Surgical-minimum patch path for the 4 MCP gaps + isolated patches folder + Phase 0/1 prep deliverables built on top — chosen and shipped this session. snowflake-mcp v22.0.6 ships Gap 1 (drop CORTEX from DESCRIBE AGENT) + Gap 2 (YAML default for describe_semantic_view_v2). tableau-mcp v6.0.6 ships Gap 4 (new get_workbook_runtime_binding tool + cumulative-history annotation on legacy tool). Gap 3 reclassified harness-side. Patches NOT git-committed because both repos contain unrelated in-flight upgrade-cycle modifications; extracted instead to mcp-patches/ folder with apply-procedure README. Phase 0/1 prep delivered alongside: 31-KPI parity baseline harness, canonical module-rule extractor (27 Sales / 16 CRM rules), 96-query verified-queries regression with 20/76 pass/fail breakdown.

Rationale

Each gap fix verified live against the original failure case. H1 (DESCRIBE AGENT works) confirmed via live execute_query. H2 (YAML form ~10x compact) confirmed via 25.7 KB inline payload. H3 (Gap 3 is harness-side) confirmed via contract_enforcer.py inspection showing G3 caps rows not bytes. H4 (additive new tool preserves contracts) confirmed via Welcome Call Dashboard test showing cumulative=PROD+DEV vs runtime=PROD only. Not-committing chosen because git diff shows 14 unrelated files in snowflake-mcp + 5 in tableau-mcp from a separate upgrade cycle — bundling would harm reviewability and risk forcing the maintainer to either reject or accept unrelated changes together. Phase 0/1 prep built in same session because the YAML scoping artefacts from the EA stream were already accessible and the operator needs runnable scripts for tomorrow. Quantitative deltas vs handoff (Sales 27 rules not 26 with 21 inter-view divergent, CRM 16 not 22, queries 96 not 64 with 21 percent pass rate, 100 percent of failures unqualified-table) surfaced for CTO so the 7-week plan can be re-scoped before Phase 1.

Alternatives Rejected

Outcome

Pending

Graph View

Backlinks