How to Connect AI Agents to Discord
With Two-Way Communication
That Actually Works

The Story — Why I Built This

Two weeks ago, I didn't know what a terminal was. I'm 38, I drive Uber, and I had zero coding experience. Then I started building with AI.

In 14 days, I built a team of 11 AI agents — a researcher, a writer, a finance analyst, a social media strategist, a content scorer, and more. They all live in Discord. They talk to each other. They cost $0.30 a day.

My main AI brain (Claude Code, running in my terminal) sends them tasks. They deliver. The system works.

Or at least, I thought it worked.

At 4 AM on March 5th, I discovered that my AI brain could send messages to the team — but couldn't hear a single response. The team had been delivering work for hours. Nobody was listening.

This guide is everything I learned building this system, including the critical bug and the 10-minute fix. Whether you're using Claude Code, ChatGPT, or any AI tool — if you want to build a team of agents on Discord, this will save you days of trial and error.

The Architecture — How It All Connects

The key components:

• Discord Server — your team's workspace. Each agent gets a channel.
• Agent Platform — runs the AI agents. We used OpenClaw (open-source, free), but this works with any framework.
• Discord Bot — connects the agent platform to Discord. Routes messages to the right agent.
• Webhooks — let your main AI brain SEND messages to any channel.
• Listener — lets your main AI brain READ messages from all channels. This is the part everyone misses.

Step 1 — Set Up Your Discord Server

// Save webhook URLs in a JSON file for easy access
{
  "general": "https://discord.com/api/webhooks/YOUR_ID/YOUR_TOKEN",
  "research": "https://discord.com/api/webhooks/YOUR_ID/YOUR_TOKEN",
  "content": "https://discord.com/api/webhooks/YOUR_ID/YOUR_TOKEN",
  "finance": "https://discord.com/api/webhooks/YOUR_ID/YOUR_TOKEN"
}

Step 2 — Install an AI Agent Platform

You need something that runs your AI agents and connects them to Discord. There are several options:

# Install Docker first, then:
git clone https://github.com/open-claw/open-claw.git
cd open-claw
docker compose up -d openclaw-gateway

Step 3 — Connect Your AI to Discord (Sending)

Sending messages from your AI brain to Discord is straightforward using webhooks:

# Send a message to any Discord channel via webhook
curl -s -X POST "YOUR_WEBHOOK_URL" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Hey team, here is your task...",
    "username": "AI Brain"
  }'

That's it. Replace YOUR_WEBHOOK_URL with the URL you saved in Step 1. The message appears in Discord as "AI Brain" (or whatever username you set).

Sending from Claude Code / any terminal AI

If you're using Claude Code or any AI that can run terminal commands, it can send to Discord directly:

# Your AI reads the webhook URLs from a config file
# and sends tasks to the team

You: "Send SCOUT a research task about competitor pricing"

AI: [reads webhook JSON → sends curl to #research]
      "Task sent to #research channel. SCOUT will pick it up."

Step 4 — The Part Everyone Misses

Here's the question nobody asks:

"When your agents finish a task and respond in Discord... does your AI brain actually see it?"

If you're using webhooks to send messages, the answer is almost certainly no.

Webhooks are one-way. They can push messages into a channel. They cannot read messages from a channel. Your AI has a mouth but no ears.

This is exactly what happened to us. We had 11 agents. 12 Discord channels. A manager that delegated tasks. Agent-to-agent communication that passed stress tests. Everything looked perfect.

Then at 4 AM, I casually asked my AI: "Did you see MAYA's message?" And it said: "No. I can't see Discord messages."

MAYA had finished a complete analysis hours earlier. Tagged the AI brain. The message sat there. Nobody was listening.

The Broken Loop — What Went Wrong at 4 AM

Here's why this bug is so insidious: everything appears to work.

• You send a task → it appears in Discord. Check.
• Your agent responds → the response appears in Discord. Check.
• Your agent tags your AI brain → the tag appears in Discord. Check.
• Your AI brain sees the response → FAIL. Silent fail. No error. No warning.

You'd never know unless you specifically asked "did you see that?" And if you don't ask... the work piles up unseen. Your agents do their jobs. Your AI brain never gets the results. You become the manual relay — the exact thing you were trying to automate.

Step 5 — Build the Listener (The Fix)

The fix is simple: a script that polls Discord channels and writes new messages to a file your AI brain reads.

How it works

1. Script runs in the background, polls all channels every 30 seconds
2. Uses the Discord REST API (same bot token your agent platform uses — no conflicts)
3. Tracks the last message it saw per channel (no duplicates)
4. Writes new messages to an "inbox" file
5. Flags messages that mention your AI brain or come from you
6. Your AI reads the inbox at the start of every session

import json, os, time, urllib.request

BOT_TOKEN = "your_bot_token"  # Same token your agent platform uses
INBOX_FILE = "discord-inbox.md"
POLL_INTERVAL = 30  # seconds
STATE_FILE = ".listener-state.json"

# Your channel IDs
CHANNELS = {
    "general":  "123456789",
    "research": "123456790",
    "content":  "123456791",
    # ... add all your channels
}

def fetch_messages(channel_id, after_id):
    """Fetch messages after a given ID from Discord API."""
    url = f"https://discord.com/api/v10/channels/{channel_id}/messages?limit=10"
    if after_id:
        url += f"&after={after_id}"
    req = urllib.request.Request(url, headers={
        "Authorization": f"Bot {BOT_TOKEN}"
    })
    with urllib.request.urlopen(req) as resp:
        return list(reversed(json.loads(resp.read())))

def run():
    # Load last-seen message IDs
    state = json.load(open(STATE_FILE)) if os.path.exists(STATE_FILE) else {}

    while True:
        for name, ch_id in CHANNELS.items():
            messages = fetch_messages(ch_id, state.get(ch_id))
            for msg in messages:
                author = msg["author"]["username"]
                content = msg["content"]
                timestamp = msg["timestamp"][:19]

                # Write to inbox file
                with open(INBOX_FILE, "a") as f:
                    f.write(f"### [{timestamp}] #{name}\n")
                    f.write(f"**{author}:** {content}\n\n")

                state[ch_id] = msg["id"]
            time.sleep(0.3)  # Respect rate limits

        json.dump(state, open(STATE_FILE, "w"))
        time.sleep(POLL_INTERVAL)

run()

This is the simplified version. The full version (included with this guide) adds:

• [MENTIONS YOU] flags on messages that tag your AI
• [FROM YOU] flags on your own messages
• Rate limit handling (Discord limits API calls)
• Error recovery (keeps running even if a request fails)
• State persistence (survives restarts, no duplicate messages)
• First-run initialization (doesn't dump old history)

# Run it in the background
python3 discord_listener.py &

# Or add it to your startup script so it always runs

7 Gotchas From 14 Days of Building

Launch Checklist