Natural-language operations
Describe the server you want. The agent resolves your request into concrete channel, category, role, and permission changes — not suggestions, actual operations.
Discord administration, operated by an agent
Tell your server
what to become.
Ubitscord.ai inspects the live server, plans each change, asks before anything destructive, and executes through typed tools — with a record of everything it did.
Developer Hub·1,842 members
connectedCreate a professional developer community.
Thinking…
I'll add Development, Community and Staff sections, create a Developer role, and lock staff channels to moderators.
Inspect Discord server420ms
4 categories, 11 channels, 6 roles
Create category310ms
Development
Create channel280ms
#frontend
Create channel265ms
#backend
Create role340ms
Developer
Set channel permission overwriteRunning
#staff-internal — deny @everyone
- Typed operations
- 23
- Bot permissions
- 8
- Admin access
- None
Getting started
Three steps to an AI operator
01
Connect Discord
Sign in with OAuth. We read your identity and the servers you can manage.
02
Install the bot
Authorise Ubitscord.ai on one server. Seven permissions, never Administrator.
03
Describe the outcome
Open the chat and say what you want. Watch each operation execute live.
Capabilities
Built to actually operate a server
Not a chatbot that answers questions about Discord. An agent with a constrained set of real operations and a policy layer around them.
Typed Discord tools
Every action runs through a named MCP tool with a JSON schema. There is no generic API escape hatch the model can abuse.
Plans before mutations
Structural and destructive work is staged as a numbered plan bound to your account, your server, and that exact set of operations.
Server auditing
A health score with prioritised findings: exposed staff channels, administrator sprawl, duplicate roles, dormant channels.
Undo and history
Every mutation records its previous state. Reversible operations can be rolled back; irreversible ones are named as such.
Inspect, plan, execute, verify
The agent re-reads the server after mutating it instead of assuming success. Multi-step reasoning is built into the loop.
Approval model
Nothing destructive happens without you
Every operation carries a risk level. Reads and additive changes run immediately. Renames and permission changes follow your setting. Deletions always stop and wait.
- Safe
Inspect, create channels, categories, non-privileged roles.
- Moderate
Rename, restructure, change permissions, post messages.
- Dangerous
Delete channels or roles, grant privileged permissions.
Execution plan
1 destructive3 operations on Developer Hub
- 01Rename channel to #web-devmoderate
- 02Set channel permissions on #announcementsmoderate
- 03Delete channel #old-generaldangerous
Auditing
Find the problems you stopped seeing
Ask the agent to audit your server and it walks the real permission graph: staff channels @everyone can read, roles that quietly hold Administrator, duplicate roles from three moderator generations ago.
Each finding comes with the fix. Apply them one at a time or as a single reviewed plan.
82/ 100 server health
- 2 staff channels visible to @everyone
- 6 channels with no activity in 45 days
- 3 duplicate roles
- 5 structural recommendations
Security
Production-grade guardrails, not prompt instructions
Authorisation lives in backend policy code. These are properties of the system, not instructions to the model.
Server-side authorisation
The guild identity is resolved from your connection record before any tool runs. The model never supplies a guild id.
Untrusted Discord content
Channel names, topics, and messages reach the model inside a data envelope. Instructions found there are never followed.
Approval binding
An approval is bound to one execution and a hash of the exact operations. Re-sending it does nothing.
Least-privilege bot
The bot never requests Administrator and refuses to grant a role more permissions than it holds.
Encrypted credentials
Discord OAuth tokens are encrypted with AES-256-GCM at rest. No secret reaches the browser.
Complete audit trail
Every mutation records the actor, arguments, previous state, result, and execution id.
Pricing
Plans that scale with your communities
One action per agent turn, plus one per change applied to Discord. Inspecting and auditing does not cost extra.
FAQ
Questions worth asking
If you are handing an agent write access to a community, you should interrogate it first.
Does the bot need Administrator?
No. Ubitscord.ai requests View Channels, Manage Channels, Manage Roles, Send Messages, Manage Messages, Read Message History, and Manage Webhooks. It refuses to create or grant Administrator roles.
What stops the AI from deleting my server?
Destructive operations are classified DANGEROUS and cannot execute without explicit approval that names each operation. Reads and additive changes run immediately; renames and permission changes follow your settings.
Can someone hide instructions in a channel name?
Everything returned from Discord is passed to the model as untrusted data with an explicit policy that it is never an instruction. Injection attempts surface as findings.
Can I undo what it did?
Yes for reversible operations — created channels and roles, renames, permission overwrites, and posted messages record their previous state. Deletions are permanent in Discord, and the product says so.
What counts as an AI action?
One action per agent turn, plus one per change applied to Discord. Inspecting and auditing your server does not consume extra actions.
Which model runs my requests?
Each plan maps to a model tier. Model access never changes the authorisation rules — those live in backend policy code.