Performance & ROI / Churn Buster MCP
Churn Buster MCP
The Churn Buster MCP connects Claude, ChatGPT, Claude Code, Codex and other AI assistants that support remote MCP servers to your Churn Buster analytics.
Ask about recovery or cancel flows in plain language, and the answer comes from your account's data. The MCP is in beta and available by invitation.
What it can answer
- Is this change significant? It checks a move against your account's normal range and flags periods that haven't finished resolving.
- How did we do? Recovery rate by count or dollars, the four outcomes, the split between card updates and retries where your platform reports it, and the recovery curve.
- What changed? When your dunning schedule changed, each schedule's share of campaigns (including A/B tests and targeted schedules), and which billing platforms your failed payments came from.
- How are our cancel flows doing? Save rate, save rate by cancel reason, and how often each offer is accepted.
- Why doesn't this match Stripe? How your processor's dashboard counts differently, and how to line the two up.
- What does this metric mean? Every definition links back to Learn.
What it can't see yet
Filters by processor, plan or segment; individual customers or campaigns; email or SMS content and deliverability; decline codes; test-mode data; billing; campaign settings and what changed in them, beyond when your dunning schedule changed; and the text of cancel feedback.
Going deeper
Segmentation is usually where an investigation finds its answer: first renewal customers against everyone else, monthly against annual plans, one processor against another. Until the MCP can filter by segment, you can export your campaign data from the dashboard and cut it yourself (our team can add custom columns), or bring in our team.
It reads, it doesn't change anything
The MCP can't edit campaigns, settings or customers. Everyone who connects sees the same account data, and only for accounts they belong to that have the MCP turned on. If the assistant can't answer something, it can send a note to our team so we know what to build. We keep a log of MCP requests for about 30 days.
Set it up
- The account owner turns on the MCP under Settings → Integrations → MCP, or we can turn it on for you.
- Add the server URL,
https://mcp.churnbuster.io/mcp, to your assistant:- Claude (Free, Pro, Max): Settings → Connectors → + → Add custom connector, paste the URL, and leave Advanced settings empty.
- Claude (Team, Enterprise): an organization Owner adds it under Organization settings → Connectors, then each person clicks Connect.
- ChatGPT: turn on Developer mode under Settings → Security and login, then go to Plugins → + and create a developer-mode app with the URL and OAuth authentication.
-
Claude Code:
claude mcp add --transport http churn-buster https://mcp.churnbuster.io/mcp, then/mcpto sign in. -
Codex:
codex mcp add churn-buster --url https://mcp.churnbuster.io/mcp, thencodex mcp login churn-buster.
- Sign in to Churn Buster and click Approve on the read-only access screen.
Disconnect
Remove the connector in your AI app to disconnect yourself. The account owner can also turn off the MCP in Churn Buster, which disconnects everyone on the account.
Questions to start with
- "Has our recovery rate changed in a way that matters this quarter?"
- "Which outcome is costing us the most, and has that changed?"
- "Why doesn't Churn Buster match our Stripe dashboard?"
- "Which cancel reasons save the fewest customers, and which offers do customers take?"
How it measures
The MCP uses the same methodology as your dashboard. It counts complete cohorts only and all four outcomes, credits recovery no matter who retried the card, and judges every change against your account's normal range. The full methodology is public on Learn.