A Discord bot that enables AI assistants to ask questions to humans in real-time through the Model Context Protocol (MCP).

This MCP server creates a bridge between AI assistants (like Claude) and humans via Discord. When an AI needs human input, it can ask questions directly in Discord and wait for responses.

Example use cases:

• AI needs project-specific information only you know
• Confirmation before executing sensitive operations
• Gathering requirements or preferences during development
• Getting feedback on generated content
• Handling ambiguous instructions that need clarification

• Rust 1.70+ (install here)
• Discord account
• A Discord server where you have admin permissions

1. Go to Discord Developer Portal
2. Click "New Application" → Name it (e.g., "AI Assistant Helper")
3. Go to "Bot" section in left sidebar
4. Click "Reset Token" → Copy the token (you'll need this!)
5. Under "Privileged Gateway Intents", enable:
   • ✅ MESSAGE CONTENT INTENT (required!)
6. Under "Bot Permissions", select:
   • Send Messages
   • Create Public Threads
   • Read Message History
   • View Channels

1. Still in Discord Developer Portal, go to "OAuth2" → "URL Generator"
2. Select scopes: bot
3. Select same permissions as above
4. Copy the generated URL and open it in browser
5. Select your server and authorize

Enable Discord Developer Mode:

• Desktop: Settings → Advanced → Developer Mode → ON
• Mobile: Settings → Advanced → Developer Mode

Then:

• Channel ID: Right-click any text channel → "Copy Channel ID"
• Your User ID: Right-click your username → "Copy User ID"

# Clone and build
git clone https://github.com/KOBA789/human-in-the-loop.git
cd human-in-the-loop
cargo build --release

# Create test config
cat > .env.test << EOF
DISCORD_TOKEN=YOUR_BOT_TOKEN_HERE
DISCORD_CHANNEL_ID=YOUR_CHANNEL_ID_HERE  
DISCORD_USER_ID=YOUR_USER_ID_HERE
EOF

# Test the connection
export $(grep -v '^#' .env.test | xargs)
cargo run --bin human-in-the-loop

If successful, the bot should appear online in your Discord server.

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "human-in-the-loop": {
      "command": "/path/to/human-in-the-loop",
      "args": [
        "--discord-channel-id", "YOUR_CHANNEL_ID",
        "--discord-user-id", "YOUR_USER_ID"
      ],
      "env": {
        "DISCORD_TOKEN": "YOUR_BOT_TOKEN"
      }
    }
  }
}

Add to MCP settings:

{
  "mcpServers": {
    "human-in-the-loop": {
      "command": "human-in-the-loop",
      "args": [
        "--discord-channel-id", "YOUR_CHANNEL_ID",
        "--discord-user-id", "YOUR_USER_ID"
      ]
    }
  }
}

Then set environment variable before starting Claude:

export DISCORD_TOKEN="YOUR_BOT_TOKEN"
claude

1. AI asks a question: When the AI uses the ask_human tool, it sends a question
2. Bot creates thread: A new thread is created in your Discord channel (or reuses existing)
3. You get notified: Bot mentions you with the question
4. You respond: Type your answer in the thread
5. AI receives answer: Your response is sent back to the AI

You: "Create a README for my project"
AI: "I'll help create a README. Let me ask you some questions about your project."
[AI uses ask_human tool]

Discord:
Bot: @YourName What is the main purpose of your project?
You: It's a tool for managing Discord bots
[Bot returns this to AI]

AI: "Great! Let me ask about the key features..."

# Create test script
cat > test_connection.py << 'EOF'
import requests
import os

token = os.environ.get('DISCORD_TOKEN')
headers = {'Authorization': f'Bot {token}'}
r = requests.get('https://discord.com/api/v10/users/@me', headers=headers)

if r.status_code == 200:
    print(f"✅ Bot connected: {r.json()['username']}#{r.json()['discriminator']}")
else:
    print(f"❌ Error: {r.text}")
EOF

python3 test_connection.py

# Download test script
curl -O https://raw.githubusercontent.com/KOBA789/human-in-the-loop/main/test_final.py
python3 test_final.py

This will:

• Initialize MCP connection
• List available tools
• Send a test question to Discord
• Wait for your response

• Verify token is correct (no extra spaces)
• Check bot was added to server with correct permissions
• Ensure MESSAGE CONTENT INTENT is enabled

• Make sure channel is a regular text channel (not forum/announcement)
• Bot needs "Create Public Threads" permission
• Try a different channel

• Check bot has "View Channel" permission
• Verify channel ID is correct
• Ensure you're replying in the thread, not main channel

• Token might be invalid or regenerated
• Check environment variables are set correctly
• Try running with --discord-token flag directly

human-in-the-loop [OPTIONS]

OPTIONS:
    --discord-token <TOKEN>          Discord bot token (or use DISCORD_TOKEN env)
    --discord-channel-id <ID>        Channel ID for creating threads
    --discord-user-id <ID>           User ID to mention in questions
    -h, --help                       Print help information

# Clone repository
git clone https://github.com/KOBA789/human-in-the-loop.git
cd human-in-the-loop

# Build release version
cargo build --release

# Binary will be at: target/release/human-in-the-loop

• Never commit your Discord token to version control
• Use environment variables or secure secret management
• Bot token gives full access to your bot - keep it safe
• Consider using separate Discord servers for testing

Contributions welcome! Please ensure:

• Tests pass: cargo test
• Code is formatted: cargo fmt
• No clippy warnings: cargo clippy

MIT License - see LICENSE file for details

• Migration to native MCP Elicitation when standardized
• Support for multiple concurrent conversations
• Configurable timeout settings
• Web dashboard for monitoring