Skip to content
mono-agent

Slack Team Bot with MCP Tools

This playbook builds a shared Slack bot that answers when mentioned in allow-listed channels, calls a custom team MCP tool plus Read/Grep, and can post proactively with SlackSendMessage.

Who this is for

Section titled “Who this is for”

A DevOps engineer running a shared team bot.

Goal

Section titled “Goal”

A Slack Socket Mode bot, mention-triggered in allowed channels, with a custom MCP tool plus Read/Grep and the SlackSendMessage tool for proactive posts.

Features used

Section titled “Features used”

Configuration

Section titled “Configuration”

The Slack section runs in Socket Mode (effective botToken and appToken values are required); Slack’s app_mention event handles real mentions, the adapter discovers its own bot user ID/username, and the bot only responds in allowedChannelIds. Put MONO_AGENT_SLACK_BOT_TOKEN and MONO_AGENT_SLACK_APP_TOKEN in .env; the source config omits credentials. The built-in @agent /model / @agent /effort thread controls and workspace-registered /<bot>-model / /<bot>-effort channel controls automatically expose runtime.model plus any configured fallbacks, with no bot-ID or Slack-specific model config. This agent opts into a specific tool allowlist instead of the allow-all default, so the built-ins it uses and SlackSendMessage are named explicitly; deployTool comes from the declared MCP server. concurrency bounds in-flight work app-wide.

{
  "runtime": {
    "model": "anthropic:claude-sonnet-4-6"
  },
  "slack": {
    "enabled": true,
    "allowedChannelIds": ["C012345"]
  },
  "tools": {
    "allowedTools": ["Read", "Grep", "SlackSendMessage", "deployTool"],
    "mcpConfigPath": "./mcp.json"
  },
  "concurrency": {
    "maxConcurrentRuns": 4,
    "maxPendingRuns": 8
  }
}

Secrets can also come from the environment: MONO_AGENT_SLACK_BOT_TOKEN and MONO_AGENT_SLACK_APP_TOKEN. The unset mention-text policy keeps one authenticated readable @name marker in the current turn; set slack.stripMentionText: true only when legacy full stripping is required.

Steps

Section titled “Steps”
  1. Create a Slack app with Socket Mode (app token connections:write), a bot token (chat:write plus commands for slash controls, users:read so the agent can address teammates by name, and channels:history / groups:history so a mention inside a thread arrives with that thread behind it), and Interactivity & Shortcuts enabled so runtime selectors can deliver block_actions over Socket Mode. Register /<bot-username>-model and /<bot-username>-effort; Socket Mode needs no Request URL.
  2. mono-agent init --model anthropic:claude-sonnet-4-6
  3. Write mcp.json with the team’s stdio MCP server; deployTool becomes available from the server declaration (MCP tools aren’t gated by tools.allowedTools). This playbook keeps a specific allowlist, so it lists deployTool for readability, but the entry is not what enables it.
  4. Set a specific tools.allowedTools (the built-ins you want plus SlackSendMessage) and add the slack section; set concurrency bounds (note: per-channel scope). Or drop tools.allowedTools entirely to keep the allow-all default.
  5. mono-agent validate, then mono-agent start (confirm Slack is running with both tokens).
  6. Mention the bot in an allowed channel and confirm a 👀 reaction, then a final reply. If a tool runs, confirm the final reply is a new message and the temporary activity message is removed.

Smoke test

Section titled “Smoke test”
Section titled “Related”

For real-turn cache measurements and tool-definition changes, enable providers.piNative.promptCacheDiagnostics and use the offline prompt-cache artifact reader.