Discord Bot Tutorial #9 - Connecting AI (Claude & ChatGPT API) to Your Bot! (Python Example)

Hi there! This is Mojangcam 😎

In Part #8, we learned how to make your bot's messages look nicer with Embeds. This time, we're finally adding the feature a lot of you have been waiting for: connecting AI to your bot! We'll build a bot where you type /ai 질문 and Claude or ChatGPT answers you.

Your server members can ask the AI anything they like, and one bot can handle things like explaining server rules, translating, and answering coding questions, so it's genuinely useful.

※ The image above isn't a real run. I haven't topped up my API credits yet, so instead of an actual response I've attached a mock-up example image. I'll replace it with a real screenshot once I add credits!

⚠️ One thing before we start: unlike the bot itself or the libraries, an AI API is a paid service that charges small amounts based on usage. That said, for a bot on a personal server, a lightweight model costs only a few hundred won (less than a dollar) a month, so it's pretty affordable.

📌 What You Need

Python 3.8+
discord.py 2.x
The anthropic library (for the Claude API) or the openai library (for ChatGPT)
An API key: get one at platform.claude.com for Claude, or platform.openai.com for ChatGPT (you'll need to add a small amount of credit)

mipsasm

pip install -U discord.py anthropic python-dotenv

This post uses the Claude API, and at the end I'll also show you the ChatGPT (OpenAI) version.

📌 Step 1 - Storing Your API Key Safely in .env

The same goes for your bot token, but an AI API key is tied to paid billing, so a leak can be a really big deal. Always keep it separate in a .env file.

INI

# .env
DISCORD_TOKEN=봇토큰
ANTHROPIC_API_KEY=sk-ant-...

📌 Step 2 - Calling Claude with the /ai Command

Python

import discord
from discord import app_commands
from anthropic import AsyncAnthropic

ai = AsyncAnthropic()  # Picks up the ANTHROPIC_API_KEY env var automatically

@tree.command(name="ai", description="Ask the AI a question")
async def ai_command(interaction: discord.Interaction, 질문: str):
    await interaction.response.defer()  # ⭐ Key part! See the explanation below
    result = await ai.messages.create(
        model="claude-haiku-4-5",
        max_tokens=1000,
        system="You are a friendly helper on a Discord server. Keep your answers short.",
        messages=[{"role": "user", "content": 질문}],
    )
    answer = result.content[0].text
    await interaction.followup.send(answer[:2000])

Let's break it down 💡

AsyncAnthropic : discord.py runs asynchronously (async), so you must use the async client. If you just use Anthropic(), your whole bot freezes while the AI is answering!
defer() : Discord treats a slash command as failed if it doesn't get a response within 3 seconds. AI responses usually take longer than 3 seconds, so you need defer to put it in a "thinking..." state
model : The example uses claude-haiku-4-5, which is fast and cheap. If you need smarter answers, switch to claude-sonnet-5 (it costs more)
system : This sets the bot's personality and role. Change it however you like to fit your server's vibe
[:2000] : Discord messages are limited to 2,000 characters, so you need to cut the text to avoid errors

📌 Step 3 - Making It Remember the Conversation (Optional)

The code above starts a new conversation every time, so it can't remember the previous question. Let's make it remember the last 10 messages per channel.

vim

from collections import defaultdict

history = defaultdict(list)  # Conversation history per channel ID

async def ask_ai(channel_id: int, 질문: str) -> str:
    history[channel_id].append({"role": "user", "content": 질문})
    result = await ai.messages.create(
        model="claude-haiku-4-5",
        max_tokens=1000,
        system="You are a friendly helper on a Discord server.",
        messages=history[channel_id][-10:],  # Send only the last 10
    )
    answer = result.content[0].text
    history[channel_id].append({"role": "assistant", "content": answer})
    return answer

Let's break it down 💡

history[channel_id][-10:] : If you send the whole conversation, your token costs keep growing. Trimming it to the last 10 is the key to saving money!
Saving the assistant role : You also need to put the AI's answers into the history, so follow-up questions like "about what you said earlier" work

📌 Full Code (Copy & Paste OK)

Python

# discord.py 2.x / latest anthropic / Python 3.8+
import os
import discord
from discord import app_commands
from collections import defaultdict
from anthropic import AsyncAnthropic
from dotenv import load_dotenv

load_dotenv()

intents = discord.Intents.default()
client = discord.Client(intents=intents)
tree = app_commands.CommandTree(client)
ai = AsyncAnthropic()
history = defaultdict(list)

@client.event
async def on_ready():
    await tree.sync()
    print(f"{client.user} logged in!")

@tree.command(name="ai", description="Ask the AI a question")
async def ai_command(interaction: discord.Interaction, 질문: str):
    await interaction.response.defer()
    try:
        history[interaction.channel_id].append({"role": "user", "content": 질문})
        result = await ai.messages.create(
            model="claude-haiku-4-5",
            max_tokens=1000,
            system="You are a friendly helper on a Discord server. Keep your answers short.",
            messages=history[interaction.channel_id][-10:],
        )
        answer = result.content[0].text
        history[interaction.channel_id].append({"role": "assistant", "content": answer})
        await interaction.followup.send(answer[:2000])
    except Exception as e:
        await interaction.followup.send(f"⚠️ Something went wrong: {e}")

client.run(os.getenv("DISCORD_TOKEN"))

💡 Want to use ChatGPT (OpenAI) instead? Just swap out the call part like this (pip install openai, environment variable OPENAI_API_KEY).

makefile

from openai import AsyncOpenAI
ai = AsyncOpenAI()

result = await ai.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "system", "content": "You are a friendly helper."},
              {"role": "user", "content": 질문}],
)
answer = result.choices[0].message.content

The only difference is that OpenAI puts the system message inside messages too. Otherwise, the structure is almost the same.

📌 Result

Start the bot and type /ai 파이썬이 뭐야? ("What is Python?") in your server. A "thinking..." indicator shows up for a moment, then the AI's answer gets posted in the channel. If you keep asking questions in the same channel, it remembers the earlier conversation when it answers!

※ This conversation image is also a mock-up example I made because of API costs. (Not a real run)

📌 Common Errors

404 Not Found or Unknown interaction — Cause: You called response.send_message directly without responding within 3 seconds. Fix: Always call defer() first → then reply with followup.send()
authentication_error (401) — Cause: The API key has a typo, is missing, or has expired. Fix: Double-check the key in .env, and check your credit balance in the console too
HTTPException: Must be 2000 or fewer in length — Cause: The AI's answer is over 2,000 characters. Fix: Cut it with [:2000] like in the example, or split long answers into several messages
rate_limit_error (429) — Cause: Too many requests in a short time. Fix: Adding a cooldown to the command helps — we'll cover how to build cooldowns in the next part!

📌 Wrapping Up

Today we connected the Claude and ChatGPT APIs to our bot and built a real "AI bot." Use defer to get around the 3-second limit, add per-channel conversation memory, and it really comes to life. In the next part, we'll cover cooldowns and permission management to stop people from abusing the AI command. Stay tuned 🙌

📚 Discord Bot Tutorial Series

#1 ~ #4 The Basics
#5 Button Interactions
#6 Dropdown (Select Menu)
#7 Getting Input with a Modal
#8 Styling Embeds with Images and Fields
#9 Connecting AI (Claude & ChatGPT API) ← You are here!
Bonus: Free 24/7 Hosting Roundup