Use Porter’s Snowflake connector for Claude to ask about the tables you choose through a read-only role, then compare them with your ad data.
By Porter Metrics team · Last updated October 19, 2026
ClaudePorter connectedPorter / Snowflake / Trailgear Demo✓ 4 rows| Week | Revenue | Orders |
|---|---|---|
| Sep 14 | $184K | 2,310 |
| Sep 21 | $196K | 2,402 |
| Sep 28 | $212K | 2,588 |
| Oct 5 | $203K | 2,471 |
Any table you pickPorter’s Snowflake connector for Claude takes 3 steps, with no code and no MCP server object to create in Snowflake. Connect a Snowflake user in Porter whose role can read the tables. Add Porter’s MCP URL to Claude as a custom connector and ask. MCP (Model Context Protocol) is the open standard Claude uses to connect to tools.
Sign up free at Porter, pick Snowflake as the source, then fill in the connection details. Porter then lists every table that role can see, named as database, schema and table, and you pick Claude as the destination.
In Claude, open Customize, then Connectors, choose + Add, then Add custom connector. Finally, name it Porter Metrics and paste this URL:
https://mcp.portermetrics.com/mcpSign in to Porter, then try this:
PromptUsing my Snowflake table SALES.PUBLIC.ORDERS, show total REVENUE and ORDERS by week of ORDER_DATE for the last 8 weeks.
Each use case comes with a prompt to copy, the columns it uses and who it’s for. Your columns are your own, so each prompt uses example tables and uppercase column names, the way Snowflake stores them. Name one table and its date column in each prompt, then swap in yours.
One prompt gives you a working map of the table, instead of scrolling SELECT * results in Snowsight column by column.
PromptUsing my Snowflake table ANALYTICS.MARKETING.DAILY_CAMPAIGNS, list its columns as metrics or dimensions, then total each metric by month of DATE. Which months have data, and which metrics stay at zero?For: data teams onboarding a table · marketing analysts · agencies
The review opens with the weekly table already built instead of a ticket asking for it.
PromptUsing SALES.PUBLIC.ORDERS, show REVENUE and ORDERS by week of ORDER_DATE for the last 12 weeks, with the change versus the previous week. Name the 2 weeks that moved most.For: RevOps and finance · growth marketers · founders
You answer the thread with the segment and the size of its drop, before anyone opens a pivot table.
PromptUsing SALES.PUBLIC.ORDERS, compare REVENUE by CHANNEL and by REGION for the last 7 days of ORDER_DATE versus the 7 days before. Rank the segments by change.For: marketing analysts · RevOps · agencies
As a result, you get revenue per dollar of spend by day, without anyone writing the JOIN or a new view.
PromptQuery MARKETING.ADS.DAILY_SPEND and SALES.PUBLIC.ORDERS separately for the last 30 days. Total SPEND by SPEND_DATE and REVENUE by ORDER_DATE, then line them up by day and show revenue per dollar of spend.For: growth marketers · data teams
The budget talk then moves to the revenue finance already trusts, set against what paid search cost each week.
PromptFor the last 12 weeks, show Google Ads Cost by Week next to REVENUE by week of ORDER_DATE from SALES.PUBLIC.ORDERS, then revenue per dollar of Cost. Did weeks with more spend bring more revenue?For: growth and performance marketers · agencies
Porter brings the Snowflake tables you pick into Claude, next to your ad and CRM data. Need Cortex Agents, Cortex Analyst, Cortex Search or SQL you write? Use Snowflake’s managed MCP server, which needs no extra account. You can run both.
Give Porter a Snowflake user whose role can read the tables, then add one MCP URL to Claude. There’s no MCP server object or OAuth integration to create in Snowflake.
Claude reads every table your Porter role can SELECT and tells you what moved, while your Snowflake data stays exactly as you set it. You grant the role and revoke its grants anytime.
Ask about warehouse revenue next to Google Ads, HubSpot, GA4 and other Porter sources in one answer, instead of exporting either side into a sheet.
Porter reads each table’s columns and offers them to Claude as metrics and dimensions. You ask for REVENUE by week, and Claude builds the query.
The Snowflake connector comes with every Porter plan, including Free, so your first table costs nothing to try. Snowflake still bills the warehouse credits each query uses.
You can start free, since Porter’s Snowflake connector for Claude comes with every plan. Use the calculator for your number of accounts, or see Porter pricing for every plan.
$33.33/mo$6.67 per account
An account is one ad account, Stripe account, store, page or GA4 property. The Free plan has unlimited accounts with 30 days of history.
The Snowflake connector lets Claude read the rows of the tables you connect, and every column becomes a field it can ask for. There's no fixed field list, because each table brings its own columns. Claude reads one table per question and combines the results itself.
| What Porter reads | How it works |
|---|---|
| Tables | Each table is its own account, named as database, schema and table |
| Columns | Read from the table and offered to Claude as metrics (numbers) and dimensions (labels and dates) |
| Column names | As Snowflake stores them: uppercase unless the column was created in quotes |
| Not read | Files in stages, Cortex services, two tables in one query |
| What | Snowflake connector |
|---|---|
| Fields | No fixed list; each column of your tables is a metric or dimension |
| Actions | Reads only, so your database stays exactly as you set it |
| Reads | One table per question, named as database, schema and table |
| History | 30 days on the Free plan, 15 in AI chats |
| Access | Read-only role: USAGE on a warehouse, database and schema, SELECT on tables |
| Not read | Files in stages, Cortex services, two tables in one query |
| Query costs | Runs on your warehouse; a resumed warehouse bills at least 60 seconds |
Yes. Claude sees only the tables that the role you create for Porter can SELECT, and you can revoke its grants anytime. Your tables, views and data stay exactly as you set them.
Porter needs USAGE on one warehouse, USAGE on the database and schema, and SELECT on each table or view Claude should read (Snowflake's access control privileges). Grant nothing else, especially not OWNERSHIP or write privileges.
A dedicated extra-small warehouse keeps Claude's questions apart from your pipelines, so you can see its credit use separately.
Data teams usually ask 4 things first: SQL, credits, Claude Code, and which tables Claude sees. Then come 3 fixes: numbers that don't match Snowsight, a connection Snowflake refuses, and slow answers.
No. Porter reads your table's columns and hands them to Claude, so you ask for REVENUE by week and Claude builds the query. Name the table and the date column, and write column names as they appear in Snowflake, so Claude matches them to the fields Porter lists.
Yes. Every question runs on your Snowflake warehouse, and a suspended warehouse that wakes up bills at least 60 seconds, then per second (Snowflake's warehouse billing rules). Snowflake can reuse an identical query's result for 24 hours when the data hasn't changed (Snowflake's result cache).
Yes. Porter's Snowflake connector covers answers in Claude chats, while Snowflake's Cortex Code plugin brings Snowflake into Claude Code. The plugin routes Snowflake prompts to the Cortex Code CLI, which can also create objects such as dynamic tables. Porter's MCP servers for marketing data list what else Claude reads through Porter.
Claude sees the tables that the role you give Porter can read. Porter lists all of those, and Claude queries only the ones you ask about. Each table Claude queries counts as one Porter account, so a narrow role keeps both the list and the bill short.
Two Snowflake settings can make Claude's numbers different from Snowsight's. TIMESTAMP_LTZ columns show in the session time zone, which defaults to America/Los_Angeles (Snowflake's TIMEZONE parameter), so daily totals can shift. Unquoted column names are uppercase, so "Revenue" and REVENUE are different columns (Snowflake's identifier rules).
Check three Snowflake settings first. Snowflake disallows passwords for service users and asks human users with passwords for MFA (Snowflake's MFA and password rollout). A network policy's allowed list refuses every IP missing from it (Snowflake's network policies), and the role needs USAGE on a warehouse.
Ask again once the warehouse is running, or point Porter at a warehouse with room to spare. A Snowflake question is slow while Porter waits for a suspended warehouse to resume, and the first queries after that run on a cold cache. A busy warehouse also queues statements beyond its concurrency limit.
Porter adds other sources to Claude through the same MCP URL, so warehouse tables and live platform data sit in one chat. Start with the platforms your tables already copy, or the ones your team reports on next to them, or browse all Claude connectors instead:
As soon as Snowflake is connected, ask Claude which week moved, which channel moved it, and what paid search returned in warehouse revenue. Setting up the Snowflake connector for Claude takes one role and one URL. Since the free plan already includes Porter's MCP, your first table costs nothing to try.
Free plan · 14-day trial · No card needed
Marketing analysts: weekly numbers and the segment behind them instead of a SQL request.
RevOps and finance: warehouse revenue together with ad spend, in one answer.
Data teams: fewer ad-hoc tickets, behind a role you control.
→
→
→
→
→