Self-host

Self-host is free and good. Here's when hosted is worth it.

The open-source package does the Slack surface well. Hosted exists for two reasons: OAuth that doesn't expire every two weeks, and the AI brain that needs Workers AI. This page states the tradeoff without spin.

The one thing self-host can't give you

A self-host session token rotates out roughly every two weeks. Re-authing is the chore that quietly kills a daily habit. Permanent OAuth on hosted Pro is the fix — and the AI brain (slack_channel_summary, slack_extract_action_items, slack_find_decisions) is hosted-only because it runs on Workers AI.

$ npx -y @jtalk22/slack-mcp --setup > session token connected > expires in ~14 days # self-host > hosted OAuth connected (Pro) > expires: never

npm install path

npx -y @jtalk22/slack-mcp --setup

21 tools, MIT licensed, 12,782 installs since January. Full runtime and token control on your machine.

Docker path

docker pull ghcr.io/jtalk22/slack-mcp-server:latest

The public repo carries full self-host docs, release notes, and issue tracking.

Decision matrix

  • Self-host — $0: 21 tools, local transport control, MIT licensed, no monthly plan. Workflow-profile primitives ship in OSS. Session token rotates ~every 2 weeks.
  • Cloud Free: $0 (no card) — the AI brain with real quotas: 2,000 requests + 25 AI tool calls per month.
  • Cloud Pro: $19/mo — unlimited requests + unlimited AI tools + permanent OAuth + 2 workspaces.
  • Cloud Team: $49/mo flat for 5 workspaces + shared workflow profiles.
  • Safeguard: $199/mo (waitlist) — agent approval gates, workspace memory, audit log. All in development; join the waitlist.

When to self-host

  • You want full control over runtime and tokens
  • You prefer npm or Docker over a managed endpoint
  • You're comfortable operating the transport and re-auth yourself
  • You don't need the hosted AI brain

When to stay hosted

  • You want OAuth that doesn't expire on you every two weeks
  • You want the AI brain — summaries, action items, decisions
  • You want Claude first and Gemini CLI second without building your own remote transport

Team still deciding?

Send the workspace context to deployment review. That keeps plan fit, support expectations, and security questions on the hosted surface instead of scattered across email.