Before you start
You need Admin or Analytics access.
You need an AI tool that supports remote MCP connectors, such as Claude. To connect, see How to connect your AI tool to Skio Data MCP.
What is the Skio Data MCP?
Skio Data MCP lets you connect your own AI tool to your Skio subscription analytics, so you can ask plain-language questions and get answers based on your store's data. It uses the same calculations that power your Skio Dashboard analytics.
MCP (Model Context Protocol) is the standard that lets an AI tool securely connect to an outside data source like Skio. When you ask a question, your AI tool picks from a fixed set of approved analytics questions, runs it against your store's data, and uses the results to answer you.
Skio Data MCP is read-only. It can answer questions about your data, but it can't change subscriptions, issue refunds, or modify anything in your account.
What can I ask with Skio Data MCP?
A few metrics always use a fixed time period, no matter what dates you ask for. These are noted in the lists below.
You can ask about the same metrics you see in your Skio Dashboard analytics, grouped into the categories below.
Every question can cover a date range of up to 1 year. Some categories also let you narrow a question to one segment or one product, using the name as it appears in your Skio Dashboard. Your AI tool can combine metrics from several categories into one answer, like a weekly report comparing revenue, churn rate, and dunning recovery rate against the prior week.
See all available metrics
Overview and revenue
Store-wide only. These can't be narrowed to a segment or product.
Daily subscription counts (active, paused, cancelled, and failing) and revenue split into recurring, first-time, and one-time
Daily monthly recurring revenue (MRR)
New vs. lost subscriptions per day
New vs. lost subscribers (customers) per day
Payment Recovery (dunning)
Can be narrowed to a segment, a product, or both.
Dunning recovery rate (always covers the last 6 months)
Subscriptions entering and exiting dunning per day
Dunning entries and exits by failure reason, like expired card or insufficient funds
Number of payment retries at dunning entry and to exit dunning
Detail of individual failed-payment episodes
Subscriptions failing payment right now
Cancellations and Cancel Flow
Can be narrowed to a segment, a product, or both.
Cancellations per day
Cancellations by reason given in your Cancel Flow
Customers who started your Cancel Flow, whether or not they cancelled
Cancel Flow outcomes: saved vs. cancelled
Detail of individual Cancel Flow sessions
Churn rate for a period (one number for the whole period, so ask month by month for a trend)
Products
Can be narrowed to a segment, a product, or both.
Summary of subscription activity across all products
Subscribers, orders, and revenue by product
Breakdown by variant and subscription type: prepaid, Build-a-Box, one-time upsell, and Surprise & Delight
Billing interval distribution, store-wide and by product (for prepaid subscriptions, this is billing frequency, not shipping frequency)
Cohorts, retention, and LTV
Can be narrowed to a segment, a product, or both.
Subscription and subscriber cohort summaries
Retention by signup month, for subscriptions and subscribers, including lifetime value (LTV) by cohort
Retention by order number (2nd order, 3rd order, and so on), for subscriptions and subscribers
Store-wide subscription and subscriber LTV, including 3-, 6-, and 12-month LTV (one number for your whole store, not by cohort)
Forecasts
Can be narrowed to a segment, a product, or both.
Projected daily revenue, orders, and units from subscriptions scheduled to bill
Revenue forecast by product
Projected new subscriptions and cancellations
Recent subscription actions (cancels, pauses, skips, and date or frequency changes) compared with a baseline, to explain forecast changes
Build-a-Box
Store-wide only.
Build-a-Box summary: total quantity ordered, average quantity per box, and average distinct products per subscriber
Subscribers, orders, and value by box
Active Build-a-Box subscriptions per day
Size of changes customers make to box contents
Build-a-Box LTV
Most popular products in boxes, and one-time add-ons to box orders
Segments
Can be narrowed to a segment.
Summary metrics for one segment, or all segments
How skips, pauses, cancels, and date or frequency changes affect revenue, AOV, and order count within a segment
Winbacks
Store-wide only. These metrics cover your Winbacks journey, so they return zeros if you haven't set one up.
Winbacks summary
Daily winback recoveries
Winback performance by original cancellation reason, offer, and product
Detail of individual winback events
Volume discounts and Surprise & Delight
Store-wide only.
Active subscriptions on a volume discount and their average order value
Surprise & Delight performance and retention impact (always compares the last 30 days with the 30 days before)
Loyalty credits
Store-wide only.
Orders split by credit spent, credit earned, and recurring vs. one-time, per day or for the whole period
Credit balance changes by how credit was applied, per day or for the whole period
Total unspent credit and how many customers hold a balance (current total, regardless of dates)
Credit discounts used vs. unused, per day or cumulative
Credit expiring at the end of the month, if credit expiration is turned on
Redeemers vs. customers who've never redeemed, redemption rate, days to first redemption, and whether redeeming relates to staying subscribed (all-time, regardless of dates)
Cancel Flow results for the loyalty screen
Customers who ordered, spent credit, or earned credit in the period
Loyalty tiers
Store-wide only.
Daily tier membership, including members with an active subscription
Orders by tier, as a total or per day
Combined credit and tier redemption summary
Are there limits on what I can ask?
Yes. Every question runs within a few fixed limits that keep answers fast and prevent runaway usage. Most questions never come close to them, but they explain why a long or broad question might get cut off.
Limit | What it is | What it means for you |
|---|---|---|
Date range | Up to 1 year (366 days) per question | For anything longer, ask separate questions, like one for each year. |
Results returned | Up to 500 rows per question | A full year of daily data always fits. Broad questions over a long history, like cohort retention for your whole store, can hit this limit. Narrow the date range or ask about one segment or product. |
Response time | 60 seconds per question | Most answers come back in a few seconds. If a question times out, ask a narrower version of it. |
Usage | 60 questions per minute and 1,000 questions per day, per store | These limits are designed to catch runaway loops, not regular use. One prompt can run several questions behind the scenes, like a comparison or a multi-metric report, and that still stays well under the limit. |
How do I get the most accurate answers?
Specific questions get more accurate answers. Your AI tool decides which approved question to run and how to interpret the results, so vague wording can lead it to pull the wrong data.
Include a date range: If you don't give one, the connector defaults to the last 30 days. Say "October 1 to October 31" instead of "recently."
Name segments and products exactly: Use the segment or product name as it appears in your Skio Dashboard. The connector looks up that name, so exact names return better results.
Spell out every condition: If you want cancellations caused by failed payments, say so. "How many subscribers churned due to failed payments in the last 30 days?" is clearer than "How many subscribers churned?"
Ask multi-part questions: You can ask for several metrics at once, like revenue, AOV, churn rate, Cancel Flow save rate, and dunning recovery rate in a single weekly report.
Why does my AI tool's answer differ from my dashboard?
The data itself is the same. Skio Data MCP uses the same calculations as your Skio Dashboard analytics, so differences usually come from how your AI tool interpreted the question.
For example, if you ask "How many customers churned due to dunning in the last 30 days?" and your AI tool misses the "dunning" part, it may return all cancellations from that period. The result would be higher than your dashboard's dunning numbers, even though the underlying data is correct.
If numbers don't match:
Ask your AI tool which data it pulled and what date range and filters it used.
Rephrase the question with a specific date range, metric, and any segment or product names.
Compare the result against the same view in Skio Dashboard > Analytics, using the same date range.
Skio doesn't control your AI tool and can't see the questions you asked or the answers it gave. AI tools can sometimes choose the wrong question or misread results. When in doubt, use your Skio Dashboard as the source of truth.
How is my data kept secure?
Every connection is limited to your store's data and is read-only.
One store per connection: Skio's servers decide which store a question belongs to, based on the connection you approved. A second, independent check confirms this before any data is accessed, so your AI tool can't reach another store's data.
Approved questions only: Your AI tool can only run questions from Skio's approved set. It can't write its own queries.
Usage limits: Each store has daily and per-minute usage limits that prevent runaway activity. Normal use won't come close to them.