mailinapp
Documentation menu

AI connectors

Connect Claude, ChatGPT or Gemini to your MailInApp account and design, send and read the results of interactive emails from a chat — "build me a product-drop email," "who answered last week's survey?," "send Tuesday's newsletter to the active-subscribers list." The assistant edits your real projects and reads your real data through the same tool-use core that powers the in-studio AI copilot; a connector is just a second front door onto it, held by an AI platform on your behalf instead of the studio's own side panel.

One URL, everywhere

https://mailinapp.com/api/mcp

Every platform connects to this exact URL — it's a remote MCP server over Streamable HTTP. There's nothing to configure before pasting it: the server advertises how to authenticate (OAuth 2.1, with your explicit consent) the moment a client connects, and the platform handles the rest. Per-platform steps are in Claude, ChatGPT and Gemini.

What happens when you connect

  1. You paste the URL above into Claude/ChatGPT/Gemini's own connector settings.
  2. The platform opens MailInApp's consent screen in your browser, signed in as you.
  3. If you belong to more than one company (your own account plus any team you're an active member of), you pick which one this connection acts as — a connection is always scoped to exactly one account.
  4. You choose which permissions ("scopes") to grant — see Scopes & permissions. Everything read-only is pre-checked; sending mail is not.
  5. Approve, and the assistant can now call MailInApp's tools on that account's behalf.

Nothing is shared until you complete this flow — connecting doesn't hand over a password or an API key, it hands the platform a scoped, revocable token.

What it can do

  • Design — read and edit a project's block tree and theme, browse the 44-block catalog, apply a saved or community template, render a live preview.
  • Read results — campaign opens/clicks/responses/revenue, form question rollups, orders, send history, recurring schedules, plan usage.
  • Manage audience — list contact lists, look up or add/update contacts, manage suppressions.
  • Reply — mailbox threads and support tickets, both fenced as untrusted recipient-authored content (see Security).
  • Send — a two-step prepare_send → confirm_send flow that always surfaces the recipient count and a rendered preview before anything goes out. See Tool reference for the full list, and Security for exactly how a send is gated.

What it deliberately can't do

No tool touches Stripe/billing (a chargeback refund stays dashboard-only), mints API keys, rotates a webhook secret, changes SMTP/payments/domain settings, or manages team invites — the same carve-outs an editor teammate is already blocked from everywhere else. campaigns:send is never granted by default, and even when it is, a send still needs the two-step confirmation above.

Plan availability

Connectors are available on every plan, including Free — reading your projects and results, and building new ones, costs nothing you haven't already budgeted for. What's metered is the number of tool calls per calendar month (200 on Free, scaling up by plan) and, separately, how many recipients a connector is allowed to mail per month (0 on Free — sending needs at least Starter). See pricing for exact numbers, and Connections in the dashboard for your month-to-date usage.

Managing a connection

Connections in the dashboard (/dashboard/connections) lists every assistant you've approved — client name, granted scopes, when it last made a call — with a one-click Revoke. Revoking takes effect immediately: every token that assistant is holding stops working on its next call, no waiting for an expiry.