Discord Builder

Manage your Discord server by chatting with AI.

Warning

Beta. Discord Builder makes real, live changes to your Discord server. Only OpenAI has been tested against its live API so far. Anthropic, Google, xAI and Meta are implemented and pass mocked tests, but haven't been run against the real services yet. Try it on a test server first, and please open an issue if something breaks.

Describe what you want in plain English, like "make a private staff area for moderators" or "post a rules embed through a webhook". The AI picks the right Discord actions, runs them through your bot, and shows you every step. Anything destructive waits for your approval.

Features

• Chat to build. A ChatGPT-style interface with saved chats, streaming replies, and a row for every Discord action, which you can expand to see its input and result.
• Covers most of the Discord API. Channels, roles, permissions, members, moderation, messages, embeds, webhooks, threads, emojis, invites, events and the audit log, plus a raw API tool for everything else.
• Approval for destructive actions. Deletes, kicks, bans, timeouts and purges pause until you click Approve or Deny.
• Server profiles. Give each server a description, tone of voice, brand colors, logo and house rules. The AI follows them in every chat for that server.
• Multiple bots. Connect several Discord bots and switch between them from the sidebar.
• Five AI providers. OpenAI, Anthropic, Google, xAI and Meta. Pick a model per chat and switch mid-conversation.
• Discord previews. Embeds and webhook messages render the way they'll look in Discord.
• Light and dark themes that follow your system.

Provider support

Provider	Models	Status	Get a key
OpenAI	GPT	✅ Tested	platform.openai.com
Anthropic	Claude	⚠️ Untested	console.anthropic.com
Google	Gemini	⚠️ Untested	aistudio.google.com
xAI	Grok	⚠️ Untested	console.x.ai
Meta	Muse Spark (Meta Model API)	⚠️ Untested	ai.developer.meta.com

Untested means the integration is written to each provider's documented API and passes mocked end-to-end tests, but hasn't made a live call. Reports and fixes are welcome.

Meta retired its Llama API in July 2026. The Meta option uses its replacement, the Meta Model API. Llama models are still available through third-party hosts, which aren't built in yet.

Quick start

Requirements: Node.js 22 or newer, a Discord account, and an API key from at least one provider above.

git clone https://github.com/ToyuAU/discord-builder.git
cd discord-builder
npm install
npm run dev

Open http://localhost:5173. On first run, a setup screen walks you through:

1. Connect an AI provider. Paste an API key. It's checked before it's saved.
2. Add your Discord bot.
   1. In the Discord Developer Portal, create an application.
   2. Open Bot, click Reset Token, and paste the token into the app.
   3. On the same page, turn on Server Members Intent. It's needed to list and search members.
3. Invite the bot to your server using the generated link. It requests Administrator so the AI can make any change you ask for.

That's it. Pick your server in the sidebar and start chatting.

Configuration

You can configure everything in the app or through environment variables.

In the app. Open Settings from the bot menu at the bottom of the sidebar. Keys, bots and server profiles are saved to data/settings.json. The file is readable only by your user, ignored by git, and never sent to the browser.

With .env. This is handy on a server or for a quick start. Copy .env.example to .env:

Variable	Purpose
OPENAI_API_KEY	OpenAI key
ANTHROPIC_API_KEY	Anthropic key
GEMINI_API_KEY	Google Gemini key
XAI_API_KEY	xAI key
META_API_KEY	Meta Model API key
DEFAULT_PROVIDER / DEFAULT_MODEL	Default model for new chats, e.g. anthropic / claude-opus-5
DISCORD_BOT_TOKEN	A bot to load at startup (add more in the app)
DISCORD_GUILD_ID	Server selected by default
PORT	API port (default 3001)
MAX_AGENT_STEPS	Max model turns per message (default 40)

If a key is set in both places, the one saved in the app wins. Restart the server after editing .env.

Server profiles

Open a server's profile from the server menu, or click the server name at the top right. Everything in it is added to the AI's instructions for every chat in that server, including chats already open.

Field	Example
About	"A community for indie game developers to share progress and find playtesters."
Tone of voice	"Friendly and upbeat, a little playful. Emoji sparingly."
Brand colors	Primary #5865F2, Accent #F2A93B, used for embeds and roles
Logo URL	Used for webhook avatars and embed thumbnails
Naming conventions	"Channels are lowercase with an emoji prefix, e.g. 📢-announcements."
Always / never	"Never ping @everyone without asking first."

Anything you ask for in chat takes priority over the profile.

What the AI can do

Area	Actions
Server	Overview snapshot, edit settings and icon, audit log, scheduled events
Channels	List, create (any type, including private), edit, delete, reorder, permission overwrites
Roles	List, create, edit (replace or add/remove permissions), delete, reorder
Members	List, search, add/remove roles, nicknames, voice move/mute; kick, ban, unban, timeout
Messages	Send with embeds and link buttons, edit, read history, delete, purge, pin, react
Threads	Create threads and forum posts
Webhooks	List, create, edit, delete, post with a custom name and avatar, edit webhook messages
Emojis & invites	List, create, delete
Anything else	discord_api_request: call any Discord REST v10 endpoint (automod, stickers, onboarding…)

Safety and security

• Approvals are on by default (Settings → General). Destructive actions and any raw API call that changes something wait for your OK.
• Actions show up in the audit log. When the AI gives a reason for a change, it's recorded in your server's audit log.
• Local only. The API server listens on 127.0.0.1 and has no authentication. Don't expose it to the internet: anyone who can reach it can control your bots and spend your API credits.
• Mind the bot's permissions. Administrator is the simplest setup but also the most powerful. Scope the invite down if you only need certain features.
• Secrets stay local. .env and data/ are git-ignored. Don't commit them.

How it works

Browser (React)  ──SSE──▶  Local server (Express)  ──▶  AI provider (tool calling)
                                   │
                                   └──────────────▶  Discord REST API (your bot)

1. Your message goes to the local server, which sends the AI the chat history, your server profile, and a list of Discord tools.
2. The AI replies with tool calls, like create_channel. The server runs them against Discord and sends the results back to the AI.
3. That loop repeats until the AI answers in plain text. Destructive calls pause the loop until you approve or deny them.
4. Progress streams to the browser as it happens. Chats are saved as JSON in data/sessions/.

Project structure

server/
  index.ts            Express API: status, settings, profiles, chats, streaming
  agent.ts            Provider-neutral tool-calling loop with approval pausing
  prompt.ts           System prompt, including the server profile
  providers/
    responses.ts      OpenAI Responses API (used for OpenAI, xAI and Meta)
    anthropic.ts      Claude via the Anthropic SDK
    gemini.ts         Gemini via its OpenAI-compatible API
  bots.ts             Bot registry, token checks, invite links
  settings.ts         data/settings.json: keys, bots, server profiles
  store.ts            Chat storage in data/sessions/
  discord/            REST client and helpers (permissions, colors, embeds)
  tools/              Discord tools, one file per area
shared/               Types and provider info shared by server and client
web/src/              React client (Vite)
assets/               Logo and screenshots

Development

Command	Description
npm run dev	API server and web client with hot reload
npm run typecheck	Type-check server and client
npm run build	Build the client into dist/
npm start	Serve the API and built client on http://localhost:3001

Adding a Discord tool: write a tool({ name, description, parameters, run }) in server/tools/ and register it in server/tools/index.ts. Set destructive: true (or a function of the arguments) to require approval.

Adding an AI provider: implement the ModelProvider interface in server/providers/types.ts, then register it in server/providers/index.ts and shared/providers.ts. Each adapter converts the shared chat history into that provider's format. Data a provider needs to continue a turn, like Claude's thinking blocks or Gemini's thought signatures, is stored with the message and replayed only to that provider.

Known limitations

• Only OpenAI has been tested against its live API. See Provider support.
• No user accounts or authentication. It's meant for one person on one machine.
• No file uploads. Images for icons, emojis and avatars must be public URLs.
• One active run per chat. Start another chat to work in parallel.
• Listing and searching members requires the Server Members Intent.

Contributing

Issues and pull requests are welcome, especially reports from anyone trying the Anthropic, Google, xAI or Meta integrations. Please include the provider, model and any error message.

License

MIT