ChatGPT Work + Snowflake: What the Integration Can (and Can't) Do in 2026
Partly — there’s no one-click Snowflake connector in ChatGPT’s directory, but Snowflake ships an official Snowflake-managed MCP server for Cortex Agents that you add to ChatGPT as a custom connector. The server exposes Cortex Analyst (natural-language-to-SQL over your semantic views), Cortex Search, direct SQL execution, and other Cortex Agents as callable tools — all governed by OAuth 2.0 and your Snowflake RBAC roles, with the SQL tool’s read_only parameter defaulting to true. You wire it into ChatGPT through Developer Mode (Plus, Pro, Team, Enterprise, and Edu — not Free), paste the server URL, and sign in with OAuth. And like every MCP connection in ChatGPT, it runs inside a session you’re driving — between chats, nothing is watching your warehouse. The much-publicized OpenAI–Snowflake $200M partnership is mostly about running OpenAI models inside Snowflake Cortex, not a native data connector on the ChatGPT side.
Here’s what the ChatGPT Work + Snowflake integration actually does, how to turn it on, where the ceiling is, and what to use when you want warehouse-adjacent work that runs on its own.
What ChatGPT Work can actually do with Snowflake
Through the Snowflake-managed MCP server, added to ChatGPT as a custom connector:
- Ask questions of your data in plain English. Cortex Analyst turns “what was net revenue by region last quarter versus the quarter before?” into SQL against your semantic views and returns the answer — no dashboard, no hand-written query.
- Search unstructured data. Cortex Search runs semantic search over document and text columns, so “find support tickets mentioning the billing migration” works inside the same chat.
- Run SQL you approve. The SQL execution tool can run queries directly, but its
read_onlyparameter defaults totrue— writes require an admin to explicitly configure them, so out of the box this is a read-and-analyze surface, not a write-back one. - Chain Cortex Agents and custom tools. The server can expose other Cortex Agents, plus user-defined functions and stored procedures, as tools — up to 50 per server.
- Stay inside your RBAC. Every tool call flows through OAuth 2.0 and role-based access control, so the AI session inherits exactly the privileges (USAGE, SELECT, and only optionally MODIFY) of the Snowflake role you signed in with.
With ChatGPT Work (OpenAI’s enterprise agent app, launched July 2026), an agent can pull Snowflake numbers into a longer run that also touches your slides, sheets, and docs — a quarterly-metrics deck assembled from live warehouse data, say. Still a run you start.
How to set it up
This is two jobs: a Snowflake admin publishes the MCP server, then you add it to ChatGPT.
- On the Snowflake side, an admin creates a managed MCP server object, attaches the Cortex Analyst / Cortex Search / SQL tools you want it to expose, and sets up an OAuth security integration. The server lives at an endpoint shaped like
https://<account_URL>/api/v2/databases/{database}/schemas/{schema}/mcp-servers/{name}, and RBAC decides which roles can call which tools. - In ChatGPT, open Settings → Connectors → Advanced and enable Developer mode (available on paid plans — Plus, Pro, Team, Enterprise, Edu — not Free).
- Back in Settings → Connectors, click Create, name it “Snowflake,” paste the MCP server URL, choose OAuth, and enter the client ID and secret from the security integration.
- Authorize in the browser popup, then confirm with something read-only: “using Cortex Analyst, show me total orders by month for the last six months.”
The OpenAI–Snowflake partnership adds guided app templates for Snowflake, Databricks, and GitHub Enterprise on Business-plan workspaces, but the data path itself is still this OAuth-governed MCP connection.
The limits that matter
- It doesn’t run on triggers. There’s no “when last night’s load finishes, email me the revenue delta” or “alert me when daily signups drop 20%.” ChatGPT queries Snowflake only when you prompt it — a warehouse full of fresh numbers sits untouched until someone opens a chat and asks.
- Read-first by default. The SQL tool ships with
read_onlyset totrue, and Cortex Analyst/Search are retrieval surfaces. Writing back — updating a table, kicking off a pipeline — means an admin explicitly enabling it and granting MODIFY, which most data teams won’t do for a chat client. - Session-bound, even in agent mode. ChatGPT Work runs are long and autonomous but manually started and metered — workspace-agent runs moved to token-based pricing on July 6, 2026 — so a standing “watch this metric” is an errand you re-run and pay for, not a monitor.
- Response and shape limits. The managed server truncates tool responses at 250 KB, caps queries at a 600-second default timeout, and allows a maximum recursion depth of 10 — fine for analytic questions, but it’s a query interface, not a bulk-export or ETL channel.
- The follow-through stops at the chat. ChatGPT can compute the number and describe the trend; it won’t then send the summary from your mailbox, update the tracking sheet, and post it to the leadership channel as one motion.
If you want Snowflake-adjacent work that runs on its own: Carly
Warehouse work is often trigger-shaped: a nightly load lands, a metric crosses a threshold, a month closes. The moment you want the response to happen on the event — the revenue delta emailed the minute the pipeline finishes, a Slack alert when churn spikes, a Monday metrics digest built and sent before your first meeting — you’ve crossed past what a chat session is for.
That’s where Carly fits. Carly is an AI executive assistant built to act on triggers, not just answer in a session:
- Fires on events and schedules, 24/7, in the cloud. When your Snowflake job completes or a scheduled check runs, Carly queries the numbers, summarizes the change, and sends it — while your laptop is closed.
- Actually sends and updates. Carly drafts and sends email across Gmail and Outlook, updates records, tasks, and your CRM, and records meetings — the follow-through that stops at the chat with ChatGPT.
- Ties the warehouse to the rest of your stack. Turn a query result into an email to the owner, a row in a tracking sheet, and a post in the right Slack channel, in one flow.
- Builds the workflow by interviewing you. Tell Carly “every morning, pull yesterday’s signups and revenue from Snowflake and email me the delta versus last week” in plain English; it interviews you and builds it — no prompt engineering.
Snowflake connects to Carly with your own credentials: paste your key or connection details on carlyassistant.com/integrations and Carly can do whatever that access allows, on the same OAuth/RBAC footing Snowflake expects. Carly connects to 200+ tools across 40+ categories natively — see integrations. AI agents start at $35/month, and steps in a workflow that don’t use AI run free and unlimited.
ChatGPT Work vs Carly
| ChatGPT Work (Snowflake MCP) | Carly | |
|---|---|---|
| Ask questions of warehouse data | Yes, in a session | Yes |
| Run SQL / Cortex Analyst queries | Yes (read-only by default) | Yes, via your access |
| Write back to Snowflake | Only if an admin enables it | Yes, if your access allows |
| Reacts to a finished load or a metric change | No | Yes, on the trigger |
| Morning metrics digest, on schedule | No | Yes |
| Emails the summary to owners | No | Yes (Gmail + Outlook) |
| Updates a sheet / CRM / posts to Slack in one flow | No | Yes |
| Runs without a session open | No (agent runs are started + metered) | Yes (cloud, 24/7) |
| Setup | Admin publishes MCP server + ChatGPT Developer Mode | Describe it in plain English |
| Pricing | Paid ChatGPT plan; agent runs metered | AI agents from $35/mo |
ChatGPT’s Snowflake connection is a data copilot you steer in a chat. Carly is an assistant that acts on your warehouse events while you’re in meetings.
Frequently Asked Questions
Does ChatGPT Work with Snowflake?
Yes, for querying — but not through a one-click connector. Snowflake ships an official managed MCP server for Cortex Agents that exposes Cortex Analyst, Cortex Search, and SQL execution as tools. You add it to ChatGPT as a custom connector via Developer Mode, authenticate with OAuth, and then ask questions of your data in plain English inside a session.
Is there an official ChatGPT connector for Snowflake?
The official piece is on Snowflake’s side: the Snowflake-managed MCP server. There is no native Snowflake entry in ChatGPT’s built-in connectors directory — you add Snowflake’s MCP server yourself as a custom connector through ChatGPT Developer Mode. The OpenAI–Snowflake partnership is primarily about running OpenAI models inside Snowflake Cortex, not a consumer-facing data connector.
Can ChatGPT write to or update Snowflake?
By default, no. The managed MCP server’s SQL execution tool sets read_only to true, and Cortex Analyst and Cortex Search are retrieval tools. Writing requires a Snowflake admin to explicitly enable write mode and grant a role with MODIFY privileges — something most data teams won’t do for a chat client. For scheduled, write-capable warehouse work, you’d use a trigger-based assistant like Carly.
Can ChatGPT alert me when a Snowflake metric changes?
No. MCP tools only run inside a conversation you start — there are no event triggers, so ChatGPT won’t watch a load or flag a metric spike on its own, and ChatGPT Work’s agent runs are manually started and metered. For “when the nightly load finishes, email me the revenue delta” or “alert the team when churn crosses a threshold,” you need an assistant like Carly that runs on triggers in the cloud around the clock.
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."
