AI agents · MCP

Run QuickGrow from the assistant you already use.

Connect Claude — or any assistant that speaks the Model Context Protocol — to your QuickGrow account. It can look up contacts, read a campaign report, draft a campaign and switch an automation on. It cannot send anything to your customers unless you say yes to that exact send, in the moment.

Get started Prefer to write the code?

30 toolsOne-time approval per capabilitySending needs a second yes

mcp · a real exchange

→ get_campaign_report
  { "brand_id": "brd_7", "campaign_id": "cmp_31" }

← { "sent": 1240, "opened": 402, "clicked": 96 }

→ send_campaign
  { "brand_id": "brd_7", "channel": "EMAIL",
    "campaign_id": "cmp_44" }

← requires_confirmation: true
  "This will send “Eid Offer” to about
   1,240 recipients. Show the user first."
  confirm_token: “…”  (10 minutes)

01The tools

Thirty tools. The same ones the product runs on.

There is no cut-down agent API. Each tool calls the service that powers the equivalent screen in QuickGrow, so what an assistant sees is what you see — the same brands, the same permissions, the same data.

Look things up

The brands this connection can reach
Contacts, and one contact in full
Segments — including a preview of who matches, before anything is built

Make things

Create a contact
Create a segment from rules
Create a lead

Run things

Send a campaign (confirmation required)
Schedule one for later (confirmation required)
Cancel a schedule — the undo, and it needs no confirmation

See how it is going

Forms, and the submissions they collected
Landing pages, with their views and conversions
The dashboard overview

02Permission

Reading is cheap. Sending is not.

Every tool call clears four checks before it runs: the scope the connection carries, a per-minute rate limit, a one-time approval you gave for that capability, and — for anything that changes something — an audit entry naming the client that made the call.

Scopes are chosen when you connect: read, write contacts, write campaigns, send campaigns, control automations, write leads. A connection that never asked for the send scope can never gain it.
	Reading	Creating a draft	Sending or scheduling
A scope on the connection	Yes	Yes	Yes
A one-time approval you give per capability	No	Yes	Yes
A fresh yes for this exact campaign	No	No	Yes
Written to your audit trail, naming the client	No	Yes	Yes
Calls allowed per minute, per connection	60	10	10

The stop

An assistant cannot quietly message your customers.

Ask it to send a campaign and the first call does not send anything. It returns a preview — the campaign name, its status, the subject, the from address and roughly how many people would receive it — plus a one-time token. The assistant is instructed to show you that preview and wait. Only a second call carrying the token actually sends.

The token is bound to that connection, that action and that campaign — it will not work on another
It expires after ten minutes; after that you get a fresh preview instead of a send
Scheduling works exactly the same way
Cancelling a schedule needs no confirmation, because stopping a send is never the dangerous direction

send_campaign · step two

→ send_campaign
  { "brand_id": "brd_7",
    "channel": "EMAIL",
    "campaign_id": "cmp_44",
    "confirm_token": "…" }

← { "id": "cmp_44", "status": "SENDING" }

# audit trail
  mcp.send_campaign  campaign cmp_44
  via mcp · client: Claude Desktop

04Setting it up

Four steps, and no key to paste.

The connection is a standard sign-in, not a secret you copy around. You approve it as yourself, with your own permissions, and you can take it away in one click.

Point your assistant at QuickGrow

A conformant client discovers the server, registers itself and starts the sign-in on its own. There is no API key to generate or store.

Sign in and pick what it may do

You log in and choose the scopes: read, write contacts, write campaigns, send campaigns, control automations, write leads. Leave out what you do not want.

Approve each capability once

The first time the assistant reaches for something new it stops and points you at Settings, Connected apps, to approve that one capability. After that it just works.

Watch the trail — or pull the plug

Every change shows up in your audit trail tagged with the tool and the name of the client. Revoke the connection, or a single capability, whenever you like.

The QuickGrow public API reference: the endpoint navigator beside the specification, served from the product itself.

Standards, not lock-in

Nothing proprietary to learn.

Authorisation is an ordinary OAuth code flow with PKCE and dynamic client registration, published for discovery at .well-known so a client can find it without being told. The tools themselves are served over the Model Context Protocol. If your assistant speaks those two things, it can connect today — and if you would rather write the code yourself, the same jobs are on the REST API.

Authorisation, token, registration and revocation endpoints, all discoverable
A bearer token on every call, with the scopes checked per tool
Same account, same brands, same permissions as the portal

06Deliberate limits

The things it is not allowed to do.

Some of these look like missing features. They are decisions. An assistant is very good at doing what you asked slightly wrong, at speed, and the cheapest place to stop that is at the tool boundary.

Campaigns arrive as drafts

The create tool only ever produces a draft. An assistant can build the whole campaign — audience, template, subject — and still cannot be the thing that decides to send it.

Contacts can be created, not rewritten

There is no edit tool and no delete tool. An assistant cannot quietly change a customer record or remove one, however confidently it thinks it should.

One brand per call

Every call names the brand it is for, and runs against that brand only. The separation is enforced by the database underneath, not by the tool asking nicely.

07Questions

What people check before connecting an assistant.

Which assistants can connect?

Any client that speaks the Model Context Protocol and can complete an OAuth sign-in — Claude, for example. Nothing about the server is specific to one vendor: the authorisation server and the tool endpoint are both discoverable, so a conformant client configures itself.

Can it send a campaign without asking me?

No, and there are two separate locks. The connection has to carry the send scope, and you have to have approved that capability once at Settings, Connected apps. Then every individual send or schedule stops and returns a preview with a one-time token that names the campaign and roughly how many people would get it. The token lasts ten minutes and works only for that campaign and that action.

Can I make it read-only?

Yes. Connect with the read scope alone and every write tool refuses with a message saying which scope it would have needed. That is a good way to start: let it answer questions about your campaigns for a week before you let it touch anything.

An assistant connects as you, not as an admin account. It inherits your membership and your brands, and nothing more — so what it can reach shrinks and grows with your own access.

Signs in as a person Writes are logged

Give an assistant the keys. Keep the switch.

Connect it read-only in an afternoon, watch what it gets right, then hand over one capability at a time.

Get started AI that writes the copy