Connect your AI agent to Discord through a bot you control

You get 208 typed tools across 31 categories, a target-bound dry run before any complete guild build, and an independently verifiable readback once a build finishes. The project is pre-1.0, so validate a release in a private test server before you point it at a production community.

What you get#

• Caller-owned bot, not a hosted service. Your bot, your Discord permissions, your process. discord-mcp never asks for a separate account and never proxies your token anywhere.
• Typed contracts for 208 operations. Messaging, moderation, application commands, events, webhooks, voice, monetization, and more, each with a generated input/output schema you can inspect before you call it.
• Observe → Approve → Act → Verify for full builds. guild_blueprint_plan returns a read-only preview, you approve the exact target and plan, guild_blueprint_apply executes with checkpointed resume, and guild_blueprint_evidence reads Discord back to confirm the result matches.
• A safety boundary you can reason about. 31 of the 208 tools require both an operator-set environment flag and a per-call confirmation before they run. The rest execute like any other write tool, so you decide the bot's Discord permissions with that in mind.

Start from your goal#

• Connect for the first time. Follow the quickstart to create a bot, install the CLI, wire up your client, and send your first message.
• Get a verified Discord outcome. Turn one request into a reviewed guild build with a dry-run plan, human approval, resumable apply, and Activity Evidence.
• Operate a community day to day. Worked recipes cover raids, webhooks, multi-step pipelines, rich announcements, channel summaries, and Gateway notifications.
• Find the right tool. Browse all 208 tools by category - messaging, moderation, guild and community management, and commands and interactions.
• Run it in production. Configure resilience, telemetry, and audit logging for a deployment that outlives your test server.
• See how it is built. Read the system design - the middleware chain, the confirmation gate, and where the boundary between "typed contract" and "your bot's Discord permissions" actually sits.
• Watch a real build. An 87-second recording of an agent turning a fresh guild into a complete gaming community.
• Have a specific question? Check the FAQ.

Know the safety boundary before you connect#

discord-mcp cannot do more than your bot can do in Discord, so your bot's roles and channel overrides are the real access boundary, not a setting in this project. Tool annotations describe risk, but ordinary writes such as messages_send can execute on the first call. Only the 31 tools explicitly marked Confirmation required use the two-part __confirm and MCP_DRY_RUN gate.

Before you point this at a production server:

• Grant your bot only the permissions the workflow actually needs.
• Restrict enabled categories with MCP_CATEGORIES if your agent should not see the full surface.
• Know which of your planned mutations are confirmation-gated and which are not.
• Treat the audit trail as an operational aid, not a compliance guarantee.

Start with the quickstart or go straight to configuring a safe baseline.