Slack MCP (korotovsky’s fork) is one of the two well-maintained community Slack servers since Anthropic archived the official one. It covers the standard send + read + search surface plus DM and thread management — the right pick if you want stable, well-documented Slack agent access.
What it produces: structured tools for post_message, search_messages, read_thread, manage_dm, add_reaction. Thread context preserved so the model can reply in-thread without losing the parent message.
Best for: small teams running ops automation on Slack — incident bots, daily standup digesters, “summarize what happened in #user-feedback this week” jobs. Also a clean choice for workspace-history extraction (research, retrospective analysis).
Skip if: you specifically need the modern “post-on-behalf” patterns or richer admin APIs — those are sometimes better in Slack MCP (zencoder fork). The two forks have minor surface-area differences; check both READMEs before committing.
Setup gotchas: like all Slack MCPs, the OAuth dance is the friction. Required scopes: channels:read, channels:history, chat:write, users:read. Token in SLACK_BOT_TOKEN. The bot must be invited to every channel it should see — there’s no workspace-wide auto-grant.
Real-world workflow: a “what happened on Slack this week” agent runs Friday afternoons. It pulls the last 7 days from #engineering, #shipped, #user-feedback, summarizes by channel, posts the digest to #leadership. 5 minutes per week of agent time, replaces a 30-minute manual write-up.
Compatible alternatives: Slack MCP (zencoderai) — different fork, also active. Discord MCP for Discord-first teams, Gmail MCP when the conversation lives in email instead.
Either fork is fine. Pick one, stick with it.