Not just signup. The invite, the receipt, the failed payment. Your AI finds every email your app sends and tests each one with throwaway users. Flowproof grades what arrives, then deletes every test account, with a receipt.
Watch the 30-second tourFor apps built by prompting or by typing. No test code. No junk accounts in your inbox.
In claude.ai, Claude Desktop or ChatGPT, add a custom connector, sign in and press Allow. No API key. Claude Code, Cursor and Codex take one snippet. Then paste the prompt for your door. Your agent learns the rules from Flowproof itself: never a real inbox, a session before any user, every link used, cleanup at the end.
https://mcp.flowproof.dev/mcpNo browser in the chat? You do the clicking. Your typing isn’t counted against your app’s email speed.
Test the emails it already sends, and find the ones it forgot.
Say to your agent
“Use Flowproof to test my app’s signup and password reset emails. Map every email my app sends first. Clean up the test users when you’re done, then show me the report.”
Then, when you’re ready
Plan the emails it needs from day one, then build and test each one as you go.
Say to your agent
“Use Flowproof to plan the emails my app needs (it’s a SaaS with teams and Stripe billing), save them to the email map, then help me build and test each one as we go. Add Flowproof’s email rules to CLAUDE.md.”
Then, when you’re ready
Temp inboxes exist. Nobody owned the whole loop, from the list of emails to the last test user deleted.
01 · Map
Your agent reads your code, or plans with you, and lists every email and who gets it.
02 · Test
It signs up with throwaway addresses, triggers each email and uses the link or code inside. Each person in a flow gets their own address.
03 · Grade
Flowproof checks what arrived and what never did, then lists the fixes, worst first. Your agent can read it. So can you.
04 · Clean up
Every test account comes out of WordPress, Supabase, Stripe or anything behind a webhook, with an itemized receipt. On every plan.
Every email gets the basics. Then each kind is held to what it must say. A check without evidence says so and never lowers the grade. In claude.ai the result is a card in the chat, and each step to a better grade has a Do this button.
| What it checks | |
|---|---|
| Every email | Arrival time; SPF, DKIM and DMARC; broken, http and staging links; leaked merge tags like “Hi ,”; a missing plain-text part; duplicates. |
| Sign-up, reset, magic links and codes | Links that work once and then expire, a wrong code refused, and the “your password was changed” notice every reset owes. |
| Invites | Who invited them to what, a link to accept, and when it expires. Checked in the invitee’s inbox, not yours. |
| Notifications and digests | A link to the exact item, not the home page. A way to turn them off. Never an empty digest. |
| Receipts, invoices and failed payments | The amount and currency, a receipt to download, the due date, and a button to update the card. |
| Trials and security notices | When the trial ends and what happens next; times with a timezone; what to do if it wasn’t them. |
| Your auth provider’s stock emails | Supabase, Firebase, Cognito and WordPress defaults that never name your product. The report includes replacements, written for that provider’s template settings. |
2 emails with nothing to fix; 2 checks carried forward from earlier runs.
Already A+. Nothing left to do for the top grade.
No fixes needed. Nothing we tested failed.
Open in dashboard →
Share
Run xmr7wr
Deleted from DrawRight: 1 user, 1 profile
Our own app, DrawRight, in one week
F56 / 100First run
A+100 / 100A week later
What changed: branded emails. Every link proven to work once and refused the second time. The reset link refused after an hour. Test users deleted, with a receipt.
Your agent reads your code for every place mail leaves the app: your email provider, your auth provider’s templates, Stripe’s receipts, the scheduled job behind the weekly digest. Each goes on the project’s email map, with who gets it.
Each run marks what it tested. The map shows every email as planned, built, tested or failed, and “14 of 38 tested” tells you what nobody has checked. Only a finished run can mark an email tested, so the number means something.
When one person’s action emails another, each gets a test address: the invite is checked in the invitee’s inbox, the approval in the owner’s. All of them are cleaned up.
Email map · Trailhead
14 of 38 tested · 6 planned, 18 built, 14 tested
| Goes to | Status | |
|---|---|---|
| Confirm your email | Goes to new user | Tested |
| Password reset | Goes to account owner | Tested |
| Your password was changed | Goes to account owner | Failed never sent |
| You’re invited to Trailhead | Goes to invited teammate | Tested |
| Payment failed | Goes to account owner | Built, not tested yet |
| Your trial ends in 3 days | Goes to account owner | Built, not tested yet |
| Weekly digest | Goes to each member | Planned |
…and 31 more, grouped by kind on the project’s page.
Most apps launch with signup, a reset and whatever their auth provider sends by default. Tell your agent what the app does before it writes any email. Flowproof plans the rest and holds each one to the bar the report grades on.
Tell your agent what the app does: accounts, teams, billing, trials. plan_emails turns that into the emails it needs, who gets each, what sends it and what it must say. They go on the map as planned.
check_sender_domain reads your domain’s DNS and gives the exact SPF, DKIM and DMARC records to add for Resend, Postmark, SendGrid, SES, Mailgun or Brevo.
Drafts for 17 kinds of email, from the welcome to the failed payment, as HTML with a text part or as React Email. Each is written to pass its own checks.
A short block for CLAUDE.md, AGENTS.md or your editor’s rules, so your agent builds every new email the same way and tests it before calling the task done.
The map moves from planned to built to tested. Emails that wait on a clock, like a trial-ending notice, can be sent now for one test user through a test hook.
check_sender_domain domain "acme.com", provider "resend" 2 of 4 ready: add DKIM and DMARC. SPF ready SPF at send.acme.com authorises your provider (include:amazonses.com). DKIM missing add TXT resend._domainkey.acme.com p=<the public key Resend shows> DMARC missing add TXT _dmarc.acme.com v=DMARC1; p=none; rua=mailto:dmarc-reports@acme.com Return path ready send.acme.com sends bounces back to Resend.
Every test user is born on a Flowproof address, and adapters get the exact list the session minted. They act on those and nothing else. No pattern deletes, no wildcards. Always a dry run first, and an itemized receipt.
| Adapter | What it does on teardown |
|---|---|
| WordPress / WooCommerce | The companion plugin flags test users, hides them from admin lists and Woo reports, and deletes their users, orders and meta. |
| Supabase / Postgres | A small Edge Function in your project (or a connection string) deletes the run’s rows from the tables you name, then from auth.users. |
| Stripe | Deletes the run’s customers, which ends their subscriptions, in test mode or live. Never refunds a payment. |
| Generic webhook | We POST the session's addresses to your endpoint. You send back a receipt. Covers anything custom. |
Coming next: Clerk, Auth0, Firebase Auth, MailerLite, Mailchimp, Kit, HubSpot, Resend, Postmark.
Completed 2026-09-19 14:02 · dry run reviewed by agent first
12 removed
Email breaks quietly. An API key expires, a DNS record changes, a plugin update drops the verification email. These catch it.
Record a passing flow once and Flowproof replays it every day. It tells you when the email stops arriving, slows down or changes. On Flow Watch and Agency.
A GitHub Action runs one of your watches against each pull request’s preview deploy on Vercel, Netlify or Cloudflare Pages, and posts the grade on the pull request. On Flow Watch and Agency.
Trial ending, the weekly digest, the day-30 check-in. A small endpoint in your app sends one now, to one Flowproof test user, and refuses anyone else.
Your server asks Flowproof whether a buyer is a live test user and skips payment only for them, so your agent can test the receipt and everything after. On Agency.
What a run proves counts for 30 days. A reset link that expired properly last week still counts today, unless the email changed since, and the report names the run that proved it.
No login. Sign up for your own app with a temp address and get a public score card on every email that arrives in the next ten minutes.
We’re The Sky Floor, a small web agency: Alan and Joel Miller, twin brothers who have built and looked after sites and web apps for clients since 2008. Their signup, verification and password-reset emails kept breaking without anyone noticing: a plugin update, a DNS or SMTP change, an expired API key. Checking them by hand meant fake accounts piling up in our clients’ databases. So we built a tool our AI coding agent could use to test those flows and clean up after itself. On its first real run, on a client’s staging site, it found a password reset that never told the account owner their password had changed.
You test hard for a few weeks, then you ship. The pricing matches that.
Cleanup is included on every plan, including free. Every test account a run creates is removed from the systems you connect, with an itemized receipt. No plan limit stops a run once it starts, or its cleanup.
| Plan | Price | What you get | Choose a plan |
|---|---|---|---|
Freeno cardEnough to test a real project properly. No card, no trial clock. | $0 |
| Create a free account |
Flow Passno auto-renewPay once while you build. We won't charge you again unless you ask. | $29once · 30 days |
| Get a Flow Pass |
Flow WatchYour flows pass. We keep checking them after launch. | $15/ month |
| Start Flow Watch |
AgencyOne caught broken signup on a client site pays for the year. | $99/ month |
| Start Agency |
Flow Watch and Agency can also be billed yearly, with two months free: $150 and $990 a year.
Create a free account, connect your AI, and ask it to test every email your app sends. Or grade your emails first, no account needed.