Build a Discord Bot Agent

Combine an Aether Forge agent with the Hermes MCP server to build a Discord bot that responds with LLM-driven decisions.

Architecture


User → Discord → Hermes Agent → MCP server → Aether Forge agent

Hermes Agent is a messaging gateway that exposes Discord (and 14 other platforms) as MCP tools. Your agent declares Hermes as an MCP server and calls messages_send / messages_list tools.

1. Install Hermes


curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes login discord
hermes accounts list

2. Create the agent


forge generate-fast \
  --name discord-helper \
  --idea "Discord bot that answers crypto questions using market data" \
  --strategy-file strategy.md \
  --planner-mode openrouter \
  --planner-model anthropic/claude-sonnet-4 \
  --output ./discord-helper

3. Wire Hermes as MCP

Edit discord-helper/aether-forge.json:


{
  "planner": {
    "mode": "openrouter",
    "model": "anthropic/claude-sonnet-4"
  },
  "mcp_servers": {
    "hermes": {
      "command": "hermes",
      "args": ["mcp", "serve"]
    }
  }
}

4. Strategy: respond to mentions

discord-helper/strategy.md:


# Discord Crypto Helper
 
## Mission
Monitor my Discord channel #crypto. When someone mentions @me or asks a question
about ETH/BTC/SOL prices, respond with current market data.
 
## Workflow each tick
1. Call `messages_list` with platform=discord, channel=#crypto, since=last_tick
2. For each new message that mentions me OR contains "?":
   - Identify the token mentioned (ETH, BTC, SOL)
   - Call `cap-market-btc-price` to get current data
   - Compose a helpful response
   - Call `messages_send` to reply
 
## Rules
- Never send more than 5 messages per minute
- Never respond to bots
- If price data is unavailable, say "data not available right now"
- Keep responses under 280 characters
 
## Memory
- Track which messages I've already responded to (don't double-reply)
- Remember user preferences (e.g., "@user prefers prices in EUR")

5. Run


forge run ./discord-helper --mode paper --auto-approve --interval 30 \
  --planner-mode openrouter --planner-model anthropic/claude-sonnet-4

The agent now polls Discord every 30 seconds and responds to mentions.

6. Verify with `forge doctor`


forge doctor ./discord-helper/aether-forge.json


[  ok] MCP server [hermes]: 10 tools available (stdio)

Adding cost control

If responses use a paid LLM, cap daily spend. Edit aether-forge.json:


{
  "x402_budget": {
    "max_per_call_usd": 0.01,
    "max_session_usd": 1.00,
    "max_daily_usd": 5.00
  }
}

The framework tracks LLM cost in x402_state.json and refuses calls past the cap.

Adding personality

Add to strategy.md:


## Voice
- Friendly but concise
- Use emojis sparingly (one per message max)
- Quote actual numbers, not "high" or "low"
- Sign off with "—" if the conversation feels done

Production deploy

Wrap Hermes credentials and OPENROUTER_API_KEY as secrets
Use the generated Dockerfile
Set --health-port 8080 and monitor /ready
Add a kill switch alert: if forge halt triggers, page someone


docker run -d \
  -e HERMES_TOKEN=$HERMES_TOKEN \
  -e OPENROUTER_API_KEY=$OPENROUTER_API_KEY \
  -p 8080:8080 \
  -v $(pwd)/.ows:/app/.ows \
  -v $(pwd)/memory.db:/app/memory.db \
  discord-helper

Variants

Telegram trading alerts — same pattern, swap discord → telegram in strategy
Slack ops bot — respond to /incident slash commands
Whatsapp DCA reminder — daily ping when it’s time to buy