Build AI Slack & Discord Bots with Agent Skills

Four-Step Event-Driven Workflow

The following workflow describes the complete lifecycle of a bot interaction, from the initial platform event through to the user-facing response. This pattern works for both Slack and Discord with minor configuration differences.

Step 1: Event Trigger

Configure the Slack MCP or Discord MCP server to connect to your workspace or guild. For Slack, enable Socket Mode in your app settings to avoid needing a public endpoint during development:

// ~/.claude/settings.json
{
  "mcpServers": {
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": {
        "SLACK_BOT_TOKEN": "$SLACK_BOT_TOKEN",
        "SLACK_TEAM_ID": "$SLACK_TEAM_ID"
      }
    },
    "discord": {
      "command": "npx",
      "args": ["-y", "mcp-discord"],
      "env": {
        "DISCORD_TOKEN": "$DISCORD_TOKEN"
      }
    },
    "llm-router": {
      "command": "npx",
      "args": ["-y", "mcp-llm-router"]
    }
  }
}

Step 2: Intent Classification

When the agent receives an incoming message event, the first step is classifying the user\u0027s intent. Instruct the LLM Router MCP to map common intents to handlers before the bot goes live: "Given a user message in our engineering Slack channel, classify it as one of: deploy-status, incident-report, tech-qa, or general-chat. Route deploy-status and incident-report to a fast model with tool access. Route tech-qa to a slower, higher-quality model with documentation search. Route general-chat to a brief canned response."

Step 3: Action Dispatch

Based on the classified intent, the agent dispatches the appropriate action. For a deploy-status query, it calls the CI MCP tool to fetch the latest pipeline state. For an incident report, it creates a Slack thread, pings the on-call engineer, and posts the incident details in a formatted block. For a tech-qa query, it searches the documentation index and returns a cited answer. Each action is completed through MCP tool calls, keeping the logic composable and auditable.

Step 4: Response

The agent composes the final response and posts it through Slack MCP or Discord MCP. For Slack, use Block Kit JSON for rich formatting — buttons, sections, and contextual links. For Discord, use embeds with colour coding (green for success, red for incidents) and reaction emoji for quick acknowledgement. The Notification MCP Skill handles any secondary notifications — for example, sending a PagerDuty alert in parallel with the Slack message for high-priority incidents.

Use Cases

AI messaging bots built with MCP skills serve a wide range of engineering and community management scenarios. Here are four concrete examples with worked prompts.

Engineering On-Call Bot

Deploy to an #incidents channel: when an alert fires, the bot automatically creates a dedicated thread, posts the alert payload in a formatted block, pings the on-call engineer based on the current PagerDuty rotation, and asks "Is this a known issue?" The team responds in the thread and the bot logs the incident timeline for the post-mortem. All of this runs through Slack MCP and Notification MCP Skill with no custom backend code.

Developer Q&A Bot

Connect to your internal documentation index: when a developer asks "how do I configure the payment service?" in #engineering, the bot classifies the intent as tech-qa, searches the Notion docs via Notion MCP, and responds with a direct answer and a link to the source page. Questions that cannot be answered from documentation are escalated to a senior engineer via a Slack DM.

Deployment Status Bot

Integrate with your CI/CD system: when someone types "/deploy status" in any channel, the bot queries the GitHub Actions MCP or CI MCP for the latest run on the main branch and responds with a formatted embed showing the build status, commit hash, and duration. Failed builds trigger a proactive notification in #deployments before anyone has to ask.

Discord Community Moderation Bot

For open-source project servers: the bot monitors specified channels for questions, classifies them as bug reports or general questions, and routes them to the appropriate thread. Bug reports are automatically cross-posted to GitHub via GitHub MCP with the message content as the issue body. General questions receive an AI-generated answer with citations from the project documentation.

Comparison Table

Use this table to choose the right skills for your bot stack. The key decision criteria are whether you need Slack or Discord (or both), whether you need event-driven triggers or polling, and whether you need multi-model routing for cost optimisation.

Frequently Asked Questions