SaaS Analytics MCP: Ask Claude About Your MRR and Churn

Most SaaS founders already use an AI assistant every day. Very few have connected it to the numbers that actually run the business.

A SaaS analytics MCP closes that gap. It gives Claude — or any MCP-capable client — read-only access to your subscription metrics, so questions like "which plan is bleeding customers?" get answered from your real Stripe or Paddle data instead of your best recollection.

This guide covers what a SaaS analytics MCP does, how to connect one in about two minutes, which tools it exposes, and the questions that are genuinely worth asking once it is running. It is written for founders and operators of subscription businesses, not for engineers building integrations.

What is a SaaS analytics MCP?

A SaaS analytics MCP is a server that exposes your subscription metrics — MRR, churn, subscribers, customers, and revenue breakdowns — as read-only tools an AI assistant can call. MCP is the Model Context Protocol, an open standard for connecting AI clients to external systems. Applied to SaaS analytics, it means your assistant queries your revenue data itself rather than waiting for you to paste it in.

The distinction from a generic database connection matters. A SaaS analytics MCP does not hand your assistant a pile of raw transactions and hope for the best. It exposes calculated metrics — MRR derived from paid invoices, churn per period, revenue concentration across your top accounts — so the answers follow standard definitions instead of whatever the model improvises from raw rows.

How to connect a SaaS analytics MCP to Claude

Chartsy's server is remote and speaks HTTP, so there is nothing to install and nothing to keep running locally. The setup is four steps:

Connect your Stripe or Paddle account in Chartsy, if you have not already, and let the first sync finish. The server reads live billing data, so there is nothing to answer without it.
Add the connector in Claude. Open Settings → Connectors → Add custom connector and paste https://mcp.chartsy.app/mcp. Claude sends you to Chartsy to sign in; approve, and you are done. There is no client ID or secret to copy.
Or one command in Claude Code, which uses the same OAuth flow:
```
claude mcp add --transport http -s user chartsy https://mcp.chartsy.app/mcp
```
Then type /mcp to sign in and check the connection.
Ask a question. Your assistant reads the tool list and picks what it needs.

For clients that cannot run an OAuth flow, generate a personal token in Chartsy under Settings → Claude & MCP and pass it as an Authorization: Bearer header instead. Tokens last a year, are shown once, and can be revoked instantly from that same tab.

What the tools actually cover

Chartsy's SaaS analytics MCP exposes 18 read-only tools. Grouped by the question they answer:

Revenue. Current MRR, MRR and paying-customer count per plan per month, the same split by country, revenue over any date range grouped by plan, country, or payment method, and refund totals with refund ratio per plan.

Growth and retention. New subscribers and churn rate over any date range, at day, week, month, quarter, or year granularity, plus progress against your active revenue and growth goals.

Customers. Top customers by lifetime revenue, customers at risk of churning — defined as two or more failed payments this quarter, plus currently paused subscriptions — and what share of MRR sits in your top accounts.

Website traffic. Visits, unique visitors, and signups over time, and traffic sources with signup rate per group, broken down by source, medium, campaign, country, referrer, or landing path.

Everything at once. One snapshot tool returns the lot over a single shared time window of up to 36 months, along with derived warnings about revenue concentration, churn spikes, and refund ratios. A second returns a single day — that day's revenue and revenue by plan, new subscribers, cancellations, refunds, visits and traffic sources, each measured against the previous day and the trailing 7-day average, so "what happened yesterday?" is also one call.

Why the snapshot tool matters more than it sounds

In practice, the most common failure mode with any analytics MCP is fragmentation. The assistant makes eleven separate calls to answer one broad question, each with a slightly different time window, and the numbers do not reconcile. The snapshot tool exists specifically to prevent this: broad questions resolve in one call, over one window, with internally consistent figures.

If you notice your assistant scattering calls, tell it to use the business snapshot.

Nine questions worth asking

Once connected, the useful shift is that you stop asking for numbers and start asking for judgement:

"How is the business doing this year?" — one snapshot call, flags included.
"Which plan has the worst churn, and is it getting better or worse?"
"What share of my MRR sits in my top five customers?"
"Which of my largest customers look at risk right now?"
"Compare signup rate by traffic source for last quarter against the one before."
"Is my refund ratio worse on any particular plan?"
"Am I on track to hit my revenue goal at the current rate?"
"Draft the revenue section of my investor update from the last six months."
"Am I going to hit my monthly event limit at this pace?"

That last category — turning figures into a written artefact — is where an MCP earns its keep over a dashboard. If investor updates are the destination, the metrics investors actually want is a useful companion piece. The data and the document live in the same conversation, so there is no transcription step and no stale copy-paste.

What it will not do

Being clear about the boundaries is more useful than overselling:

No writes. Every tool is read-only. Nothing creates, edits, or deletes anything, and nothing reaches back into Stripe or Paddle.
No customer names or emails. Customers are identified by payment-processor ID only. Look the ID up in Chartsy to see who it is.
No BigCommerce. The server covers Chartsy for SaaS — Stripe and Paddle accounts — plus sites tracked with Growth. There are no order, product, or store tools.
No charts. It returns structured data; your client decides how to present it.

One thing that does cost you something: tool calls draw on your ChartsyAI question allowance, the same monthly budget as questions asked inside Chartsy. A long conversation with many tool calls uses more of it, which is another reason the broad snapshot tools are worth reaching for.

Conclusion

A SaaS analytics MCP is the difference between an assistant that speculates about your business and one that reads it. For Stripe- and Paddle-based subscription companies, that means MRR, churn, at-risk customers, revenue concentration, and traffic attribution all become things you can simply ask about — in whichever AI client you already spend your day in.

The setup is one URL and one token. The constraint worth respecting is that the numbers are only as good as the definitions behind them, which is why calculated metrics beat raw table access every time.

Connect Chartsy to Claude in two minutes →

Frequently asked questions

Can Claude read my Stripe data directly?

Not in a useful form on its own. Claude can read raw Stripe records, but MRR, churn, and LTV are calculated metrics that require consistent definitions. A SaaS analytics MCP exposes those calculations as tools, so the figures match your dashboard rather than being improvised.

Does a SaaS analytics MCP work with Paddle?

Yes. Chartsy's MCP server reads whichever data sources are connected to your account, including Stripe, Paddle Billing, and Paddle Classic. If several are connected, your assistant can scope a question to one account.

Is it safe to give an AI assistant access to my revenue data?

Chartsy's server is read-only and never sends customer names or emails — only payment-processor IDs and calculated metrics. The remaining risk sits with your AI client: check how it stores conversations and who on your team can see them.

How many tools does the Chartsy MCP server expose?

Eighteen, covering revenue, growth and retention, customers, website traffic, account and site listings, and monthly event usage. Two of them are snapshots — one for the whole business over a shared time window, one for a single day — so broad questions resolve in a single call.

Do I need a developer to set this up?

No. It is one command in Claude Code, or a short JSON entry in Claude Desktop. There is no package to install, no server to host, and no code to write.