Quickstart

This page helps you prepare a Discord bot token for the Dis.gg MCP service quickly, then connect it to common MCP-compatible clients.

1. Create and Configure Your Discord Bot #

Use Discord's official documentation:

• Create your app and bot: Building your first Discord Bot
• Open the Developer Portal: https://discord.com/developers/applications
• Direct bot settings page (token section is on this page): Bot settings

From the same official quickstart, make sure you cover:

• fetching credentials (Application ID + Bot token)
• installation link setup
• scopes and permissions
• app installation to your server

2. Invite the Bot to Your Server #

Use the official installation flow from Discord docs:

• Set up an install link
• Adding scopes and bot permissions
• Installing your app

3. Keep Your Bot Token Safe #

Discord token handling guidance (official):

• Fetching your credentials

Minimum good practice:

• never hardcode the token in source code,
• never commit it to Git,
• rotate immediately if leaked.

Even if you use the Dis.gg MCP service from Codex, Claude Code, or another MCP client, the credential used here is still a Discord bot token.

4. Use Dis.gg MCP with Common Clients #

OpenAI SDK (Responses API) #

Official references:

• Using tools
• Connectors and MCP servers

Example (allowed_tools limits imported MCP tools):

from openai import OpenAI
import os

client = OpenAI()

resp = client.responses.create(
    model="gpt-5",
    input="List my available Discord guilds.",
    tools=[
        {
            "type": "mcp",
            "server_label": "discord_mcp",
            "server_url": "https://mcp.dis.gg/v1",
            "authorization": f"Bot {os.environ['BOT_TOKEN']}",
            "allowed_tools": ["list_current_user_guilds", "get_guild"],
            "require_approval": "never",
        }
    ],
)

print(resp.output_text)

Claude Code #

Official references:

• Connect Claude Code to tools via MCP
• Headless mode (--allowedTools)

MCP server config (example .mcp.json):

{
  "mcpServers": {
    "discord-mcp": {
      "type": "http",
      "url": "https://mcp.dis.gg/v1",
      "headers": {
        "Authorization": "Bot ${BOT_TOKEN}"
      }
    }
  }
}

Tool restriction example in headless mode (equivalent to an allowlist):

claude -p "Read the last 5 messages in #general and summarize them." \
  --allowedTools "mcp__discord-mcp__get_messages_id_range,mcp__discord-mcp__get_message"

5. List Available Tools with Your Bot Token #

Use this to verify what the Dis.gg MCP service currently exposes for your current token context:

curl -sS https://mcp.dis.gg/v1/tools/list \
  -X POST \
  -H "Authorization: Bot $BOT_TOKEN" \
  -H "Content-Type: application/json" \
  --data '{}'

6. Restrict Tools Early (allowed_tools or Equivalent) #

Before letting an LLM run freely, limit what it can call:

• OpenAI Responses API: use allowed_tools in the MCP tool definition.
• Claude Code: use --allowedTools in automation/headless runs, or limit which MCP servers are configured and approved.

Start with read-only tools, then add write tools only when necessary.

7. Security Warning (Important) #

Even with a strict tool allowlist, risks remain:

• prompt injection can steer the model toward harmful actions,
• model mistakes can still produce bad tool arguments,
• over-permissive bot rights can cause severe impact.

Read these pages before production:

• Headers Reference
• Security Model
• Usage Guide

Enforce restrictions at multiple layers:

• LLM/client-side tool allowlist,
• MCP headers (X-Allowed-*, X-Target-*, X-Permission-Bits),
• Discord role/channel permissions for the bot itself.