Custom Commands V3AI authoring
Custom Commands V3Beta

Describe it. The assistant builds it.

Connect Claude or another MCP client to your server. It reads your channels, the action catalogue, the authoring guide and 28 worked examples first, then writes the flows for you. There is no editor to learn.

Build with AI

SCNX connector

You

Post a question in #general every morning at nine and open a thread for the answers. Members should be able to suggest the next question.

Claude

I read your channels, your roles and the action catalogue first. Here is what I built:

  • A module called Daily question with two flows
  • Two storage fields: the question catalogue and the count of questions asked
  • Settings for the channel and the time it posts
  • A simulated test run, so nothing was sent to your server

The flows are saved as drafts and are not running yet. Say the word and I will activate them, or open them in the editor first.

Illustration of a conversation. The exact wording depends on your client and model.

What it saved, as a draft

The assistant saves what it wrote as a draft flow. You open it in the editor, read the steps, run the simulation and switch it on yourself.

Daily question

ActiveSave

Trigger

Change

Schedule, every day at 09:00

#1

Get storage value(get-storage-value)

Read the question catalogue from{questions}

#2

Pick the next entry(array-at)

The one after yesterday's

#3

Send message(send-message)

Post the question as an embed in#general

#4

Create thread(start-thread-from-message)

Open a thread on that message

#5

Increase storage value(increment-storage-value)

Count the question as asked in{asked}

Illustration only - actual interface may differ

Read only or read and write

You decide at connect time. A connection reaches only the servers you picked, lasts as long as you said, and can be revoked whenever you want. Behind it sit 52 tools: 36 that only read, 16 that write. On a read-only connection nothing can be changed at all.

The assistant can only change what you allowed. Every change is snapshotted first, snapshots are kept for 30 days, and putting the earlier version back is one step.

What read only allows

List and read your flows and modules

Read a module's settings and what is in its stored data

Read the run history, and single runs in detail

Look up any of the 96 triggers, 691 actions and 181 parameter types

Search the docs and the example gallery

Check a flow or a module for mistakes

What read and write adds

Create and update modules, save and delete flows

Activate a flow, write the module's page, fill in its settings

Install a module from the marketplace and reload your bot

Run simulated tests

Change stored data, after showing you exactly what would change

Put an earlier version back

Test before it touches your server

A test run through the connection is always a simulation. The bot sends nothing, changes nothing and deletes nothing. You see what each step would have done, without the values it would have used.

Nothing is sent, changed or deleted

You see what each step would have done

A flow stays a draft until you activate it, and drafts never run

Nothing is written to your stored data until you agree to the change it shows you

Test run

Simulated

Schedule fired

Would run as the daily 09:00 trigger

Get storage value

Would read the question catalogue

Send message

Would post the question in the channel from config

Create thread

Would open a thread on that message

Simulated run, nothing was sent.

Illustration only - actual interface may differ

Paced, so nothing runs away

The assistant works through a long build in batches rather than rewriting your whole server in one sitting. How quickly it can put changes live grows with your plan, and the connection keeps track of that itself instead of walking into a wall.

A change your server turned down because the flow was not valid does not count against that, so fixing mistakes costs you nothing.

What the connection cannot do

This connection is for Custom Commands V3 and nothing else. Ask it for one of these and it says so rather than trying.

Configure built-in bot modules such as levels, welcome, moderation or tickets

Start, stop or change your bot, or go anywhere near its token

Touch modmail

Touch backups

Touch analytics

Touch billing

Those stay in the dashboard, where you do them yourself.

AI authoring, answered

Any client that speaks MCP. Claude is the obvious one, and other clients with custom connector support work the same way.

Want your server in the beta?

V3 is a large system and it still moves fast, so servers come in one at a time instead of all at once. Every application is read by a person, and you hear back within a few days.

The servers already in the beta decide a good part of what we build next. What you build is yours: flows and modules stay when the beta ends.