discordctl

Infrastructure as Code for Discord servers. Describe roles, categories, channels, forums and permissions in YAML, preview changes with plan, apply them safely with apply, detect drift with audit — from the terminal or through an MCP server that AI coding agents (Claude Code, OpenCode, …) can use.

$ discordctl plan --config examples/blastorama/discord.yaml

Plan for Blastorama (123456789012345678)

+ Create role: Moderator [medium risk]
+ Create role: Playtester
+ Create category: THE LAB
+ Create channel: 🐛╺╸bug-reports
~ Update channel: 📣╺╸announcements
    permissions[Moderator]: (none) → allow SendMessages, SendMessagesInThreads
↕ Move channel: 💬╺╸general
    category: (none) → COMMUNITY

Plan: 4 to create, 1 to update, 1 to move, 0 to delete.

Features

• Declarative YAML with Zod validation, line-numbered errors, presets, imports and templates — see docs/configuration.md.
• Terraform-style plan/apply: create, update, move, import and (only when explicitly allowed) delete, with dependency ordering and human-readable diffs.
• Stable identity: resources are tracked by logical key → Discord ID in a local SQLite state file. Renaming a channel in the config renames it in Discord; it is never recreated. Existing resources are adopted by name instead of duplicated.
• Permission engine: BigInt bitfields, category → channel inheritance, explicit allow/deny/inherit, effective-permission simulation, and an audit that flags dangerous setups (Administrator on members, exposed staff channels, writable announcements, locked-out moderators, conflicting overwrites, hierarchy inversions, failed expect: assertions).
• Safe by default: apply is a dry run unless confirmed, high-risk changes need extra confirmation, deletions need --allow-delete and only touch resources discordctl manages, stale plans are rejected, the last administrative role is protected, and bot tokens are redacted everywhere.
• MCP server for AI agents: read-only by default, plans are approval-bound with short expiry and a guild-state fingerprint check; there is no tool for arbitrary Discord API calls.
• Multi-server: every config, state entry, plan and lock is scoped to one guild; an optional allow-list limits which guilds can be touched.
• Official API only: bot authentication over Discord's REST API via @discordjs/rest (discord.js project), which handles rate-limit buckets from Discord's headers. No self-bots, user tokens, scraping or undocumented endpoints.

Install

Requires Node.js ≥ 22.16 (LTS 24 recommended).

npm install -g discordctl      # installs `discordctl` and `discordctl-mcp`
# or run without installing:
npx discordctl --help

From source (for development):

git clone https://github.com/paezao/discordctl && cd discordctl
pnpm install && pnpm build
pnpm link --global

No Docker, database server or other infrastructure is needed. State lives in ./.discordctl/state.db (SQLite via Node's built-in node:sqlite).

Quick start

1. Connect a bot with the guided wizard (a few minutes):

   discordctl setup --config examples/blastorama/discord.yaml   # --config is optional

   It opens the Developer Portal, validates and stores the bot token (0600 file in ~/.config/discordctl), opens an invite link requesting exactly the permissions the config needs, waits for the bot to join, saves DISCORD_GUILD_ID to .env, and re-runs doctor until the bot's role is positioned correctly. Manual alternative: docs/bot-setup.md, cp .env.example .env, discordctl login.
2. Start from a template or from your existing server:

   discordctl init --template gaming-community --output discord.yaml
   # or capture what you already have:
   discordctl export --guild $DISCORD_GUILD_ID --output discord.yaml --guild-var

3. Check, preview, apply:

   discordctl doctor   --config discord.yaml   # connectivity, permissions, hierarchy
   discordctl validate --config discord.yaml   # offline: schema + permission audit
   discordctl plan     --config discord.yaml
   discordctl apply    --config discord.yaml   # asks for confirmation
   discordctl apply    --config discord.yaml   # second run: "No changes."

Commands

| Command | Purpose | | --- | --- | | setup [--config F] [--guild G] | Guided first-time setup: bot token, invite, role position, health checks. | | init [--template T] [--output F] | Create a config from a template (templates lists them). | | login / logout | Store/remove a bot token (hidden prompt or --token-stdin). Env vars take precedence. | | invite [--config F] | OAuth2 URL requesting exactly the permissions a config needs. | | guilds | Guilds the bot can access. | | inspect --guild G | Roles, channel tree, overwrites, bot position. | | export --guild G [--output F] | Current structure as YAML (round-trips to an empty plan). | | validate --config F [--strict] | Offline validation, references, Discord limits, permission audit. | | plan --config F [--out plan.json] [--allow-delete] [--detailed-exitcode] | Read-only preview. | | apply (--config F \| --plan plan.json \| --plan-id ID) [--yes] [--allow-high-risk] [--allow-delete] [--dry-run] | Apply with confirmation, fingerprint check and post-apply verification. | | audit [--guild G] [--config F] [--fail-on high] | Live permission audit plus drift detection. | | permissions [--config F \| --guild G] [--role R] [--channel C] [--list] | Effective-permission matrix. | | import --config F [--dry-run] | Adopt existing resources into state without changing Discord. | | doctor [--guild G] [--config F] | Health checks. | | plans list/show, approve ID, reject ID | Review plans proposed by AI agents. | | state list/rm/refresh/export/restore/reset/unlock | Inspect and repair state. | | history | Local audit log of applied changes. | | schema | JSON Schema of the config format. |

Every command accepts --json for automation. Exit codes: 0 success, 1 error, 2 changes pending (plan --detailed-exitcode) or findings at/above audit --fail-on.

Using it with AI agents (MCP)

claude mcp add discordctl -- npx -y -p discordctl discordctl-mcp

The agent can inspect servers, export and edit configs, validate, audit and produce plans. Applying requires a human: either run discordctl apply --plan-id <id> yourself (read-only server, the default), or start the server with --allow-apply so the agent can call discord_apply_plan after you confirm in the client or with discordctl approve <id>. Details: docs/mcp.md.

Example: Blastorama

examples/blastorama/discord.yaml is a complete community server for a browser arena FPS: 6 roles in hierarchy, 5 categories, 14 channels named emoji╺╸channel-name, three tagged forums, read-only rules, staff-only announcements, developer-published dev updates, a hidden staff area, expect: assertions that validate checks, and suggested onboarding.

export DISCORD_GUILD_ID=<test server id>
discordctl permissions --config examples/blastorama/discord.yaml   # simulate access before applying
discordctl plan --config examples/blastorama/discord.yaml

On a server without the Community feature, the announcement channels are created as text channels (fallbackType: text) and converted in place once Community is enabled.

Documentation

• Bot creation, invitation and required permissions
• Configuration reference
• MCP server
• Security model
• Architecture
• Discord limitations and manual steps
• GitHub Actions: plan on PR, apply on merge
• Troubleshooting
• Releasing (maintainers)

Development

pnpm install
pnpm dev -- plan --config examples/blastorama/discord.yaml   # run from source (tsx)
pnpm test           # unit + integration tests against an in-memory Discord fake
pnpm typecheck
pnpm build

Tests never touch Discord. src/provider/fake.ts emulates the behaviours discordctl relies on (role hierarchy, "can only grant what you have", name normalization, positions, Community-only types, forum tag IDs) and supports failure injection for rate limits and partial failures. An opt-in test against a real throwaway server lives in tests/integration/ (see .env.example).

Contributing

See CONTRIBUTING.md. Report security issues privately as described in SECURITY.md.

License

Licensed under the Apache License, Version 2.0. See NOTICE.