Skip to content

devxoul/agent-messenger

Repository files navigation

Agent Messenger

npm platform platform platform platform platform platform platform

Your agent messages as you — not as a bot

One CLI for Slack, Discord, Teams, Telegram, WhatsApp, KakaoTalk, and Channel Talk. Credentials extracted from desktop apps or authenticated in seconds — no API keys, no OAuth, no admin approval. TypeScript SDK included.

Table of Contents

Why Agent Messenger?

You shouldn't need a bot token to send a message.

Every platform gates API access behind OAuth apps that need admin approval — days of waiting just to send a message. And even then, your agent is a bot, not you. Different name, different permissions, different context.

Agent Messenger reads session tokens from your Slack, Discord, Teams, KakaoTalk, or Channel Talk desktop app — zero config. Telegram and WhatsApp authenticate with a one-time phone or pairing code. Either way, your agent operates as you — same name, same permissions, same context. Bot tokens are fully supported too for server-side and CI/CD use cases.

  • Auto-Extract Auth — Reads tokens from Slack, Discord, Teams, KakaoTalk, and Channel Talk desktop apps. Telegram and WhatsApp authenticate with a one-time code — still under a minute
  • Act As Yourself — Extracts your user session — not a bot token. Your agent sends messages, reacts, and searches as you. Need bot mode? Bot CLIs are included too
  • One Interface — Consistent command style across 7 platforms for supported actions (e.g. message send, message search, channel list, snapshot). Learn once
  • Agent-Native Output — JSON by default for LLM tool use. --pretty for human-readable. Structured output your agent can parse and act on
  • Token Efficient — CLI, not MCP. One skill file, one shell command per action. No server to run, no tool registration. (Why not MCP?)
  • Persistent Memory — Stores workspace IDs, channel mappings, and preferences in ~/.config so your agent never asks twice
  • TypeScript SDK — Import clients directly into your app. Full type safety with Zod schemas

Installation

CLI (global install for terminal / AI agent use):

npm install -g agent-messenger

SDK (project dependency for programmatic use):

npm install agent-messenger

The global install gives you all platform CLIs. The project install gives you both CLIs and the TypeScript SDK.

This installs:

  • agent-slack — Slack CLI (user token, zero-config)
  • agent-slackbot — Slack Bot CLI (bot token, for server-side/CI/CD)
  • agent-discord — Discord CLI
  • agent-discordbot — Discord Bot CLI (bot token, for server-side/CI/CD)
  • agent-teams — Microsoft Teams CLI
  • agent-telegram — Telegram CLI (user account via TDLib)
  • agent-whatsapp — WhatsApp CLI (user account via Baileys, pairing code auth)
  • agent-whatsappbot — WhatsApp Bot CLI (Cloud API, for server-side/CI/CD)
  • agent-kakaotalk — KakaoTalk CLI (sub-device login, LOCO protocol)
  • agent-channeltalk — Channel Talk CLI (beta, zero-config, extracted cookies)
  • agent-channeltalkbot — Channel Talk Bot CLI (beta, API credentials, for server-side/CI/CD)

Agent Skills

Agent Messenger includes Agent Skills that teach your AI agent how to use each CLI above. One skill per CLI — install only what you need.

SkillPad

SkillPad is a GUI app for Agent Skills. See skillpad.dev for more details.

Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad Available on SkillPad

Skills CLI

Skills CLI is a CLI tool for Agent Skills. See skills.sh for more details.

npx -y skills add devxoul/agent-messenger

Claude Code Plugin

claude plugin marketplace add devxoul/agent-messenger
claude plugin install agent-messenger

Or within Claude Code:

/plugin marketplace add devxoul/agent-messenger
/plugin install agent-messenger

OpenCode Plugin

Add to your opencode.jsonc:

{
  "plugins": ["agent-messenger"],
}

Quick Start

Get up and running in 30 seconds:

# 1. See your workspace at a glance
agent-slack snapshot --pretty

# 2. Send a message
agent-slack message send general "Hello from the CLI!"

That's it. Credentials are extracted automatically from your Slack desktop app on first run. No OAuth flows. No API tokens. No configuration files.

Telegram Quick Start

Get up and running with Telegram in a minute:

bunx --package agent-messenger agent-telegram auth login

# Send a message
bunx --package agent-messenger agent-telegram message send <chat-id-or-@username> "Hello from the CLI!"

The CLI automatically provisions API credentials via my.telegram.org if needed. For CI/CD, set AGENT_TELEGRAM_API_ID and AGENT_TELEGRAM_API_HASH environment variables.

SDK

Use Agent Messenger as a TypeScript library. Each platform exports a typed client, credential manager, types, and Zod schemas.

Quick Example

import { SlackClient } from 'agent-messenger/slack'

const slack = await new SlackClient().login()
const channels = await slack.listChannels()
await slack.sendMessage(channels[0].id, 'Hello from the SDK!')

Credentials are resolved the same way as the CLI — auto-extracted from your desktop apps. Call .login() with no arguments for auto-extraction, or pass credentials explicitly:

const slack = await new SlackClient().login({ token: 'xoxc-...', cookie: 'xoxd-...' })

Available Imports

Import Path Client
agent-messenger/slack SlackClient
agent-messenger/slackbot SlackBotClient
agent-messenger/discord DiscordClient
agent-messenger/discordbot DiscordBotClient
agent-messenger/teams TeamsClient
agent-messenger/whatsapp WhatsAppClient
agent-messenger/whatsappbot WhatsAppBotClient
agent-messenger/kakaotalk KakaoTalkClient
agent-messenger/channeltalk ChannelClient
agent-messenger/channeltalkbot ChannelBotClient

Each module also exports its credential manager, Zod schemas, and TypeScript types:

import { SlackClient, SlackCredentialManager, SlackMessageSchema } from 'agent-messenger/slack'
import type { SlackMessage, SlackChannel } from 'agent-messenger/slack'

Manual Credential Setup

Every client supports .login() with explicit credentials for advanced use cases:

import { SlackClient } from 'agent-messenger/slack'

const client = await new SlackClient().login({ token: 'xoxc-...', cookie: 'xoxd-...' })
const messages = await client.getMessages('C01234567')

Real-time Events (Slack)

import { SlackClient, SlackListener } from 'agent-messenger/slack'

const client = await new SlackClient().login()
const listener = new SlackListener(client)
listener.on('message', (event) => {
  console.log(`New message in ${event.channel}: ${event.text}`)
})
await listener.start()

Real-time Events (KakaoTalk)

import { KakaoTalkClient, KakaoTalkListener } from 'agent-messenger/kakaotalk'

const client = await new KakaoTalkClient().login()
const listener = new KakaoTalkListener(client)
listener.on('message', (event) => {
  console.log(`New message in ${event.chat_id}: ${event.message}`)
})
await listener.start()

Supported Platforms

Feature Slack Discord Teams Telegram WhatsApp KakaoTalk Channel Talk (beta)
Auto credential extraction
Send & list messages
Search messages
Threads
Channels & Users partial
Reactions
File uploads
File downloads
Workspace snapshots
Multi-workspace / account
Activity feed
Drafts
Saved items
Unread messages
Sidebar sections
Pins & bookmarks
Scheduled messages
Channel management
Reminders
User groups
Real-time events (SDK)
Bot support

⚠️ Teams tokens expire in 60-90 minutes. Re-run agent-teams auth extract to refresh. See Teams Guide for details.

Platform Guides

Use Cases

Gathering Context

Pull context from conversations before you start working — no tab-switching, no skimming.

"Read the #incident-api-outage thread in Slack and summarize the root cause, timeline, and action items so I can write the postmortem."

"Search our Discord #architecture channel for any previous discussion about event sourcing before I write a proposal."

"Check my unread messages across all Slack channels and tell me if anything needs my attention."

"Look through #frontend in Slack for messages about the login page redesign from the past two weeks and summarize the decisions made."

"Search Teams for any messages mentioning 'API deprecation' so I know if this was discussed before."

Communicating & Reporting

Send updates, file reports, and notify your team — all from a prompt.

"Post a deployment summary to #releases in Slack with the commit hash, changelog, and deploy status."

"Send a message to the #standup channel with what I worked on yesterday, what I'm doing today, and any blockers."

"Cross-post this announcement to #general in Slack, the announcements channel in Discord, and the General channel in Teams."

"Upload the latest test coverage report to #ci-results in Slack."

"React with ✅ to the last message in #deploy-requests to confirm I've handled it."

Automation & Pipelines

Wire messaging into your CI, scripts, or agent workflows.

"After every CI run, post the build status and test results to #builds in Slack — include the branch name and commit link."

"When a long-running migration finishes, notify me in Discord with the final row count and elapsed time."

"Every morning at 9am, snapshot my Slack workspace and post a summary of active channels to #team-pulse."

"Send an alert to #oncall in Slack whenever the error rate exceeds 1% — include the service name and a link to the dashboard."

"Read the latest message in #releases, then cross-post it to our Discord announcements channel."

...and More

These are just starting points. Your agent has full read/write access to Slack, Discord, Teams, Telegram, WhatsApp, KakaoTalk, and Channel Talk — anything you'd do manually in a chat app, it can handle for you. If you build something cool with Agent Messenger, let me know!

Philosophy

Why CLI, not MCP?

MCP servers expose all tools at once, bloating context and confusing agents.

MCP Approach Agent Skills + CLI
All tools loaded at once Load only what you need
Bloated context window Minimal token usage
Agent confused by options Focused, relevant tools
Requires a running server One shell command per action

With Agent Messenger, your agent loads the skill it needs, uses the CLI, and moves on. No wasted tokens. The SDK complements the CLI for when you need programmatic access—same credentials, same platform coverage, full type safety.

Why not OAuth?

OAuth requires creating an app and workspace admin approval—days of waiting just to send a message. Agent Messenger skips all of that. Your desktop apps already have valid session tokens; Agent Messenger extracts them directly so you can start messaging immediately.

For server-side bots and CI/CD, bot tokens are fully supported via agent-slackbot, agent-discordbot, agent-whatsappbot, and agent-channeltalkbot.

Inspired by agent-browser from Vercel Labs.

Contributing

bun install    # Install dependencies
bun link       # Link CLI globally for local testing
bun test       # Run tests
bun test:e2e   # Run e2e tests
bun typecheck  # Type check
bun lint       # Lint
bun lint:fix   # Lint with autofix
bun format     # Format
bun run build  # Build

Thanks

License

MIT

About

Messenger automation CLI for AI agents. Slack, Discord, Teams and more.

Topics

Resources

Contributing

Stars

Watchers

Forks

Packages