Second Brain

snowflake

10 min read

Snowflake

Cloud data warehouse with Cortex Analyst for AI-powered CRM reporting, analytics, and 856 MCP tools.

Current State

Snowflake with session-based MCP server (856 tools). Cortex Analyst in production for CRM reporting views. MCP v21 upgrade completed. Data pipeline: Salesforce → Fivetran → Snowflake → Tableau/Power BI. AI SQL generation via Cortex Analyst Gen 1.

Key Learnings

  • snowflake-cortex-analyst-prod-status-crm-reporting-view — Cortex Analyst production status for CRM reporting

  • snowflake-mcp-v21-upgrade — MCP v21 upgrade details and patterns

  • cortex-analyst-testing-gen-1-ai-sql-generation-instructions — Cortex Analyst Gen 1 AI SQL generation testing

  • ea-standing-directives-a-mcp-registration-is-session-based — MCP registration is session-based for Snowflake

  • runwal-fy-2026-27-strategic-direction-consolidate-entire — FY 2026-27 strategic direction

  • n8n-god-agent-mcp-rebuild-v5 — N8N integration with Snowflake

  • runwal-group-business-context-indian-real-estate-developer — Runwal Group business context

  • aj-ea-core-mcp-servers-permanently-registered — Salesforce, Snowflake, Fivetran, and Tableau are permanently registered for the aj-ea workspace in ~/.claude.json unde

  • operational-mcp-servers-were-duplicated-across-three-workspa — The 4 permanent operational MCP servers (Salesforce, Snowflake, Fivetran, Tableau) were found registered in /root/aj-ea

  • mcp-duplicate-registrations-removed-from-aj-workspace — The 4 permanent operational MCP servers (Salesforce, Snowflake, Fivetran, Tableau) were found duplicated in /root/aj-wor

  • mcp-server-registrations-bloat-context-of-unused-workspaces — MCP servers registered in ~/.claude.json under a workspace path load into that workspace’s context every session, even

  • salesforce-mcp-not-permanently-registered-in-aj-ea — Despite MEMORY.md claiming Salesforce, Snowflake, Fivetran, and Tableau are permanently registered under projects./root/

  • salesforce-mcp-registration-missing-from-aj-ea-workspace — MEMORY.md claims aj-ea has 4 permanent workspace-default MCP servers (Salesforce, Snowflake, Fivetran, Tableau) but no s

  • aj-ea-workspace-mcpservers-empty-despite-memory-md-claim — The /root/aj-ea project in ~/.claude.json had mcpServers: {} (completely empty) despite MEMORY.md asserting 4 perm

  • aj-ea-snowflake-fivetran-tableau-mcp-also-missing-permanent — When investigating the missing Salesforce MCP, only Salesforce was registered. Snowflake, Fivetran, and Tableau — also c

  • aj-ea-4-permanent-mcp-servers-registered-09-apr-2026 — All 4 workspace MCP servers for /root/aj-ea are now permanently registered in ~/.claude.json under `projects./root/a

  • permanent-mcp-servers-are-exactly-8 — The definitive permanent MCP set (user-scope, both workspaces, every session) is: excalidraw, graphiti-mcp, hostinger-ad

  • autoresearchclaw-uses-academic-apis-not-generic-web-search — AutoResearchClaw queries OpenAlex, Semantic Scholar, and arXiv APIs directly for literature discovery — not generic web

  • autoresearchclaw-academic-api-advantage-over-web-search — AutoResearchClaw uses OpenAlex, Semantic Scholar, and arXiv APIs to retrieve real peer-reviewed literature with structur

  • oracle-has-zero-citation-integrity-autoresearchclaw-fills-ga — The Oracle stack currently has no citation verification capability — web search is used generically with no validation o

  • autoresearchclaw-citation-verification-4-layer-standalone — AutoResearchClaw’s citation verification uses 4 layers: arXiv ID validation → DOI resolution via CrossRef/DataCite → tit

  • autoresearchclaw-four-extractable-standalone-capabilities — AutoResearchClaw contains 4 capabilities worth extracting as standalone Oracle utilities independent of its paper pipeli

  • four-layer-citation-verification-anti-hallucination-pattern — AutoResearchClaw’s citation verification uses a 4-layer chain: (1) arXiv ID lookup → (2) DOI resolution via CrossRef/Dat

  • autoresearchclaw-standalone-capability-extraction-pattern — The highest-value AutoResearchClaw capabilities — citation verification (4-layer), claim verification, and academic lite

  • autoresearchclaw-four-capability-gaps-filled-in-oracle — AutoResearchClaw fills 4 specific gaps in the Oracle stack that previously did not exist: (1) 4-layer citation verificat

  • snowflake-v2-progressive-disclosure-for-large-skill-files — A 1230-line monolithic SKILL.md incurs full token cost on every invocation even when only one phase is needed. Solution:

  • snowflake-v2-skill-pattern-reduces-entry-point-token-cost-95 — The Snowflake v2.0 progressive-disclosure pattern splits a monolithic skill file into a lightweight entry point (~54 lin

  • desktop-bridge-snowflake-uuid-prefixed-tools-no-vps-registra — The Claude Desktop app proxies Snowflake connector tools into SSH/VPS sessions as UUID-prefixed deferred tools (`mcp__<u

  • snowflake-mcp-is-connected-false-vs-health-check-healthy — The Snowflake MCP is_connected tool can return false even when the bridge is fully functional because it checks a ca

  • snowflake-mcp-is-connected-false-despite-healthy — The is_connected tool on the Snowflake MCP connector may return false while health_check_snowflake returns `health

  • claude-desktop-bridge-mcp-tools-appear-disappear-at-session — The Claude Desktop connector proxies remote MCP servers into VPS SSH sessions as UUID-prefixed tools (e.g., mcp__25c8feb

  • desktop-connector-mcp-list-snapshot-at-session-start — Claude Desktop-bridged MCP connectors are snapshotted at session start; toggling a connector in the Desktop app mid-sess

  • 56-of-127-skills-intentionally-out-of-fusion-lint-scope — The fusion-invariants-lint only tracks delivery-intelligence orchestrators and specialists — not all 127+ skills on disk

  • fusion-lint-56-skills-intentionally-out-of-scope — The fusion-invariants-lint.py ORCHESTRATORS list tracks 71 of 127 on-disk skills. The 56 unlisted skills are intention

  • helix-full-stack-delivery-intelligence-system — # Helix — Full-Stack Delivery Intelligence System

  • redash-query-runner-extensibility-is-primary-integration-poi — Redash’s primary extension seam is BaseQueryRunner: implement run_query(query, user), get_schema(), test_connection(), a

  • rbac-rls-at-query-execution-layer-is-non-negotiable-enterpri — For internal enterprise analytics platforms handling multi-function data (Sales, Finance, HR, Projects), row-level secur

  • runwal-vantage-project-scope-canonical-reference — Runwal Vantage is a Tableau-grade internal BI platform forked from Redash. Scope: 9 Runwal functions, 500→1000 users, 5

  • vantage-snowflake-only-single-source-architecture — Runwal Vantage consumes only Snowflake as its data source. All upstream sources (Salesforce, SAP S/4, SharePoint CSV/Exc

  • internal-enterprise-platform-drops-40-percent-of-saas-archit — When a BI platform is deployed internal-only (single tenant, mandate-driven rollout, no external users), billing, multi-

  • r-dash-v1-approved-stack-react19-fastapi-cubejs-rq — R-Dash v1 architecture: React 19 + Vite 6 (SPA, auth-walled internal), FastAPI (async Python backend), RQ + rq-scheduler

  • bi-platform-single-source-principle-all-via-snowflake — R-Dash (Runwal’s internal BI platform) is architected as Snowflake-only consumption — Salesforce, SAP S/4, SharePoint CS

  • r-dash-single-snowflake-connector-zero-ingestion — R-Dash connects to Snowflake only as its sole data source. All upstream systems (Salesforce, SAP S/4, SharePoint CSV/Exc

  • redash-query-runner-extension-architecture — Redash’s core extensibility is the query_runner framework: BaseQueryRunner (run_query/get_schema/test_connection/annotat

  • r-same-snowflake-only-no-ingestion-architecture — R-Same is a consumption-only BI platform: Snowflake is the single active data source, and zero ingestion/ETL code exists

  • internal-platform-scope-eliminates-40pct-architecture-surfac — Clarifying that a BI platform is internal-only (no external users) eliminates billing, multi-tenancy, public dashboards,

  • r-same-snowflake-only-consumption-zero-ingestion — All upstream data sources (Salesforce, SAP S/4, SharePoint CSVs/Excels, other files) flow into Snowflake via existing up

  • redash-query-runner-dormant-activatable-via-env — Redash bundles ~70 query runners but only enables those listed in the REDASH_ENABLED_QUERY_RUNNERS environment variable.

  • sharepoint-snowflake-no-ingestion-in-analytics-platform — All upstream data sources (SharePoint, SAP, Salesforce) must materialize into Snowflake via dedicated pipelines (Fivetra

  • internal-platform-declaration-eliminates-40pct-architecture — Explicitly classifying a platform as internal-only (not a product) eliminates billing, multi-tenancy, marketing site, tr

  • sharepoint-must-materialize-to-snowflake-upstream — SharePoint CSV/Excel ingestion must NOT be handled inside R-Same. Any attempt to add file-polling or Graph API calls to

  • redash-query-runner-framework-extensibility-pattern — Redash’s query-runner framework uses BaseQueryRunner (run_query, get_schema, test_connection, annotate_query, configurat

  • r-same-snowflake-only-consumption-ar4-constraint — R-Same consumes Snowflake as its single data source — zero ingestion code lives inside the platform. All upstream source

  • sharepoint-sap-salesforce-all-via-snowflake-zero-ingestion — R-Same consumes Snowflake as its sole data source — zero ingestion code inside the platform. All upstream sources (Sales

  • fernet-encryption-for-data-source-credentials — Data source credentials (Snowflake username, password, account, warehouse) are encrypted with Fernet (symmetric AES-128-

  • taskflow-drizzle-missing-foreign-keys-data-integrity — The TaskFlow repo uses Drizzle ORM but defines all relational columns (teamId, assigneeId, creatorId, taskId, authorId,

  • redash-query-runner-activation-via-env-var-not-code — Redash’s 70+ bundled data-source drivers (query runners) are not activated by code changes — each driver calls a module-

  • sharepoint-data-flows-via-snowflake-not-direct-ingestion — SharePoint CSV/Excel files are NOT ingested directly by R-Same. They flow upstream via Fivetran/n8n/custom shim → Snowfl

  • r-same-repo-scaffold-and-wave-status — R-Same lives at github.com/ARJ-TECH/r-same (private), local path /opt/infra/r-same/. Wave 0 scaffold (55 files) and

  • drizzle-orm-foreign-keys-not-enforced-by-default — Drizzle ORM schema columns typed as plain int (e.g., teamId, assigneeId, creatorId) do NOT enforce referential i

  • r-same-sharepoint-snowflake-upstream-only — R-Same consumes Snowflake only — all SharePoint CSV/Excel, Salesforce, and SAP data flows upstream via Fivetran/n8n befo

  • redash-query-runner-framework-ported-wholesale-to-r-same — Redash’s query_runner framework (BaseQueryRunner, BaseSQLQueryRunner, module-level register(), REDASH_ENABLED_QUERY_RUNN

  • cube-query-uses-5min-hs256-jwt-with-security-context — Live Cube wiring authenticates via a 5-minute HS256 JWT signed per-request in execute_cube_query. The security context

  • column-masks-reapplied-on-cached-results-defense-in-depth — Column masks are re-applied on cached QueryResults even after Cube returns data. This is a defense-in-depth pattern so t

  • widget-exactly-one-source-invariant-enforced-at-three-layers — R-Dash Widget model enforces the exactly-one-source invariant (either Cube semantic query OR raw SQL, never both/neither

  • query-revision-auto-snapshot-wired-into-service-update-not-r — R-Dash M9 QueryRevision auto-snapshots are triggered inside QueryService.update() at the service layer, not in the rou

  • exactly-one-source-invariant-on-widget-requires-three-layer — Widget data source exclusivity (semantic vs SQL, exactly one active) requires enforcement at three layers: Pydantic vali

  • column-mask-reapplication-on-cached-query-results — Column masks must be re-applied to cached query results at read time, not just at query execution time. If a role is rev

  • cube-client-module-level-secret-read-breaks-prod — Reading CUBEJS_API_SECRET at module import time with a dev fallback means the dev key is baked in before runtime env var

  • column-masks-reapplied-post-cache-defense-in-depth — RLS column masks are re-applied to QueryResult cache hits at read time, not just at query execution time. This ensures r

  • drizzle-orm-fk-not-auto-enforced-requires-explicit-reference — Drizzle ORM schema definitions do not automatically enforce foreign key relationships — all FK columns (teamId, assignee

  • snowflake-catalog-gotchas-v22-0-3 — # Snowflake catalog patterns that look documented but aren’t

  • salesforce-mon-org-health-2026-04-26-3-p1-anomalies-storage — # Salesforce Mon Org Health Pull — 2026-04-26

  • expertise-pushback-pattern-honest-scope-correction-over-enth — # Expertise-Pushback Pattern

  • composition-over-installation-pattern-for-external-plugin-ev — # Composition-over-Installation pattern for external plugin evaluation

  • bible-v1918-r41-r44-retrofit-pattern — # Bible v19.1.8 R41-R44 retrofit pattern (proven on snowflake-mcp v22.0.4)

  • mcp-bible-v1919-r45r46r47-wrapper-impl-arity-discipline — # Bible v19.1.9 — Wrapper-Impl Arity Discipline (R45-R47)

  • bible-v19110-sync-protocol-r45-gap-for-all-in-one-class-mcp — # Bible v19.1.10 sync protocol + R45 applicability gap

Known Issues

(none)

Decisions

Relationships

Sources

  • Vault notes: 9 references

Contradictions

(none)

fivetran | salesforce | tableau | runwal

Graph View

Backlinks