Slack MCP (zencoderai’s fork — actively maintained) gives an agent two-way Slack access. The model can post to channels, send DMs, search recent messages, manage threads, and pull workspace history. It’s how you wire “agent → human team chat” notifications without writing Slack API code by hand.
What it produces: a tool surface covering post_message, list_channels, search_messages, read_thread, add_reaction, update_message. Full thread context preserved so the model can reply in-thread, not just blast a top-level message.
Best for: ops automation (“post the daily build status to #engineering”), customer-research synthesis (“read the last 100 messages in #user-feedback and pull patterns”), incident triage (“alert me in Slack when Sentry MCP detects a new error class”).
Skip if: your team is on Discord, Teams, or Telegram — wrong tool. Skip for one-off notifications; a curl to a webhook is faster than wiring an MCP.
Setup gotchas: Slack’s auth is the friction. You need a Slack app installed in the workspace with the right scopes (chat:write, channels:read, channels:history minimum). Tokens go in env: SLACK_BOT_TOKEN. Be careful with chat:write.public — it lets the bot post to channels it isn’t invited to, which is great for ops bots and terrible if leaked.
Real-world workflow: post-deploy hook. After every wrangler deploy, an agent reads the diff, generates a 3-line “what changed” summary, posts to #shipped. Keeps async-team in sync without me writing release notes manually.
Compatible alternatives: Slack MCP (korotovsky) — different fork, slightly different scope, also active. Discord MCP for Discord-first teams.
Use this fork over the official archived one. Maintainership matters.