Claude + Snowflake: What the Integration Can (and Can't) Do in 2026
Yes — Snowflake has an official, live listing in Claude’s connectors directory, and the relationship goes far deeper than a connector. Anthropic and Snowflake signed a multi-year, $200M partnership in December 2025: Claude models run natively in Snowflake Cortex across AWS, Azure, and GCP, and Claude Sonnet 4.5 powers Snowflake CoWork (renamed from Snowflake Intelligence at Summit 26), the “ask your data” agent for Snowflake’s 12,600+ customers. The directory connector gives Claude read-only access to structured and unstructured warehouse data on web, desktop, mobile, Claude Code, and the API. One honest caveat before you click Connect: it’s not one-click OAuth like Gmail — a Snowflake admin has to create a managed MCP server in your account first. And like every Claude integration, it only works inside a conversation you start.
Here’s exactly what the integration does, how to set it up, where the limits bite, and what to use if you want warehouse work that runs on its own.
What the Snowflake connector does
The connector talks to a Snowflake-managed MCP server — generally available and created inside your account with plain SQL (CREATE MCP SERVER plus a YAML tool spec, up to 50 tools). Depending on what your admin exposes, Claude can use:
- Cortex Analyst — governed natural-language-to-SQL over semantic views. “What was EMEA churn last quarter?” runs against the vetted business definitions, not Claude’s guess at your schema.
- Cortex Search — retrieval over unstructured data you’ve indexed in Snowflake.
- Cortex Agents and custom tools — UDFs and stored procedures your team chooses to expose, plus
SYSTEM_EXECUTE_SQLif the admin allows raw queries.
Auth is Snowflake OAuth 2.0 (recommended) or a PAT, and Snowflake’s docs include an explicit claude.ai and Claude Desktop setup path. One deprecation to know: the open-source snowflake-labs/mcp repo is deprecated — its own README points you to the managed MCP server. If a tutorial tells you to clone it, the tutorial is stale.
How to set it up
- Admin, in Snowflake: run
CREATE MCP SERVERwith a YAML spec listing the tools to expose (a Cortex Analyst semantic view is the highest-value starting point). The endpoint follows the patternhttps://<account_url>/api/v2/databases/{db}/schemas/{schema}/mcp-servers/{name}. Not available in gov regions. - Admin: set up Snowflake OAuth (or issue PATs) for the users who’ll connect.
- You, in Claude: add Snowflake from the connectors directory (or Settings → Connectors), point it at your account’s MCP server, and sign in. Remote connectors require a paid Claude plan.
- Ask questions: “top 10 accounts by consumption this month” or “summarize support tickets mentioning latency” — governed queries, your permissions.
The limits that actually matter
- Admin-mediated setup. No MCP server in the account, no connection. If you’re not the admin, this is a request to your data team, not a checkbox.
- Read-only by design. The directory connector is scoped to reading data. It won’t run DML for you, kick off a task, or fix a failed pipeline.
- No triggers, no monitoring. Claude answers inside a conversation you start. It won’t notice that revenue moved two sigma overnight, that a Task failed at 3am, or that Snowpipe has been erroring since Tuesday.
- Laptop-bound for anything scheduled. The closest thing to “running on its own” is Claude Cowork’s scheduled tasks — fixed clock, and only “while your computer is awake and the Claude Desktop app is open.” (Snowflake’s CoWork agent, meanwhile, lives inside Snowflake — it’s not orchestrating your email and calendar.)
So Claude is excellent for “ask the warehouse anything, governed, right now” — and not built for “watch the warehouse and tell me when something breaks or moves.”
If you want Snowflake work that runs on its own: Carly
The moment you want the warehouse to reach you — the anomaly email at 6am, the Monday pipeline digest, the task-failure alert — you’ve crossed past what a chat connector does.
That’s where Carly fits. Carly is an AI executive assistant built to act on triggers. Snowflake connects to Carly via your own API credentials — a key-pair for the SQL API (POST /api/v2/statements, key-pair JWT auth) — paste them on carlyassistant.com/integrations and it works like any native connection:
- Nightly anomaly watch — at 6am Carly runs the revenue-by-segment query, compares against the trailing 28-day baseline, and emails you only if something moved more than two sigma. Silence means normal.
- Monday pipeline digest — a scheduled query joins bookings and usage; Carly turns the JSON result into a five-bullet narrative with the CSV attached, in your inbox before standup.
- Task-failure sentry — each morning Carly polls TASK_HISTORY() and Snowpipe load errors, summarizes root cause, and drafts the message to the data-eng owner.
- Clean division of labor — Snowflake keeps the heavy compute on its own scheduler (
CREATE TASK,CREATE ALERT); Carly runs the queries, reads the results, and does the communicating — email, tasks, CRM, calendar. - No-code setup — describe the workflow in plain English; Carly interviews you and builds it.
- Actually sends — drafts and sends email across Gmail and Outlook, updates records, manages tasks.
- Connects to anything — 200+ native integrations, plus any other tool via your own API key.
AI agents start at $35/month, and steps in a workflow that don’t use AI run free and unlimited.
Claude vs Carly
| Claude (Snowflake connector) | Carly | |
|---|---|---|
| Ask the warehouse in plain English | Yes (Cortex Analyst, governed) | Yes, via SQL API |
| Search unstructured data | Yes (Cortex Search) | Yes, via exposed queries |
| Scheduled queries + digests | No | Yes |
| Watches for anomalies / failed tasks | No | Yes |
| Works while laptop is closed | No | Yes (cloud) |
| Emails results to stakeholders | No (Gmail draft-only) | Yes (Gmail + Outlook) |
| Setup | Admin creates MCP server, then connect | Paste key-pair credentials, describe the workflow |
| Pricing | Pro $20 / Max $100–$200 | AI agents from $35/mo |
Claude with the Snowflake connector is the best governed ask-your-data chat in the directory. Carly is the teammate that watches the warehouse and reports in on its own.
Frequently Asked Questions
Does Claude work with Snowflake?
Yes — officially. Snowflake has a live listing in Claude’s connectors directory: read-only access to structured and unstructured data via a Snowflake-managed MCP server, on Claude web, desktop, mobile, Claude Code, and the API. The prerequisite is that a Snowflake admin runs CREATE MCP SERVER in your account first — it’s not instant OAuth like consumer connectors.
What’s the relationship between Anthropic and Snowflake?
A $200M multi-year partnership announced December 3, 2025. Claude models are hosted in Snowflake Cortex across all three clouds, Claude Sonnet 4.5 powers Snowflake CoWork, and Claude drives Cortex Code (now CoCo) — which Snowflake called its fastest-growing product ever at Summit 26. Snowflake was also a launch partner in the Claude Marketplace.
Should I use the snowflake-labs/mcp open-source server?
No — it’s deprecated, with a notice pointing to the Snowflake-managed MCP server. Use CREATE MCP SERVER in your account instead; it’s GA and it’s what the directory connector expects.
Can Claude alert me when a Snowflake task fails or a metric moves?
No. The connector answers questions in a chat you start; it doesn’t poll TASK_HISTORY() or watch metrics between conversations. Carly runs scheduled queries via the SQL API, checks task and Snowpipe health, and emails the right person — 24/7, in the cloud. AI agents start at $35/month.
More: Claude + Databricks · Claude + Tableau · Claude + Power BI · Claude + Google Sheets · Claude connectors · Best AI agents for productivity
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
