Back up your Slack locally, and search it from Claude Desktop and Cursor (via MCP)

TL;DR

You can make "that thing we discussed in Slack eight months ago" reachable from Claude Desktop or Cursor — without exporting anything to the cloud:

• Back up your Slack locally (on your own machine),
• run a small MCP server over that archive, and
• connect Claude / Cursor / Codex (any MCP client) to it, read-only.

No admin approval, no OAuth app, no tokens to generate.

The problem

• Slack's free plan hides messages older than 90 days.
• Even on paid plans, your history lives on someone else's server.
• Leaving a company usually means losing your DMs.
• And when you're already working inside an AI tool, you want to pull from Slack there — not tab away to hunt for it.

What MCP is, quickly

MCP (Model Context Protocol) is a shared standard for connecting AI apps to outside data and tools. A data source runs a small MCP server; any MCP client (Claude Desktop, Cursor, Codex, …) can connect and use it. Think of a wall socket and a plug — same shape, so every client plugs into the same thing.

The flow

1. Back up Slack locally with Empowia
2. Enable Empowia's MCP server
3. Register it in Claude Desktop / Cursor
4. Ask questions that pull from your Slack

Steps

1. Back up your Slack locally

Grab Empowia from the download page and run it. It reuses your already-signed-in Slack session — no admin approval, no OAuth app, no tokens — and pulls the conversations you can already see into a local database. Your Slack sign-in is stored encrypted at rest via Windows DPAPI.

Back up the workspace/conversations you care about first: the MCP server serves this local archive, so it needs to exist before you connect a client.

2. Enable the MCP server

In the app, open Settings → MCP and turn the server on. A small local service now runs on your machine that can answer questions about your backed-up archive (nothing is uploaded).

The same screen shows a config block with a copy button. The app generates it for your specific install, so copy it from there rather than hand-writing it.

The shape of it (the app fills in the real command path for your install):

{
  "mcpServers": {
    "slacksync": {
      "command": "(the executable path shown in Settings → MCP)",
      "args": ["--mcp"]
    }
  }
}

3. Register it in Claude Desktop / Cursor

Add an MCP server in your client and paste the block you copied.

• Claude Desktop: merge it into claude_desktop_config.json under mcpServers
• Cursor: add it in the MCP section of settings

Save, restart the client, and Empowia's tools show up.

4. Ask questions that pull from Slack

Now just talk to your AI the way you already do. When your question touches Slack, the client reaches through MCP to read the relevant parts.

"How did we implement rate limiting on the export endpoint last year — and were there any gotchas the team flagged?"

It pulls the relevant threads and answers in context.

The screenshot above is Empowia's built-in Ask. MCP brings that same "search and answer with citations" experience to external clients like Claude and Cursor.

What you can do (10 read-only tools)

Empowia exposes read-only tools only:

• search_messages — full-text search
• list_workspaces / list_channels / list_files
• get_message — a single message
• collect_thread — a whole thread
• collect_recent_messages / collect_messages_about
• list_todos / list_decisions — extracted to-dos and decisions

There are no write tools — it never posts to or changes anything in Slack.

Why local matters

• The archive stays on your machine — no Empowia cloud, account, or telemetry.
• The MCP server is local too — you choose which clients connect, and can remove them anytime.
• The only thing that leaves your computer is whatever your AI client sends to its own provider when you ask. Empowia isn't in that path.

Caveats

• Windows 10/11 only for now (Mac/Linux is a notify-me on the site).
• It's not code-signed yet, so Windows SmartScreen warns on first run (More info → Run anyway).
• The free version is the full app capped at 20 conversations; a one-time purchase (no subscription, offline unlock code, no account) removes the cap.

Wrap-up

Your Slack history is one of the best records you have of how your team actually works — and it's most useful right inside the AI tools you already use. MCP bridges the two while keeping the archive on your own machine.

• Tool: Empowia for Slack
• Download: https://empowia.com/download

Feedback very welcome — especially on what would make you trust a closed, local-first tool like this.

Comments

[Mads Hansen]

Local-first is a useful privacy property, but it is not the same as authorization. I would be careful framing “no admin approval” as a benefit: technically avoiding an OAuth app does not grant permission to retain company messages or DMs after access ends.

For an enterprise-safe version, I would want:

• workspace-level consent and a visible archive scope
• retention/expiry rules plus verified secure deletion
• revocation and offboarding that purge data no longer authorized
• preservation of channel/DM ACLs in the local index
• localhost-only MCP binding with client authentication
• encrypted database, attachments and backups—not only the Slack session
• audit records for archive, search and message retrieval
• citations carrying workspace/channel/message/time provenance

“Read-only” limits writes back to Slack, but it does not limit disclosure into an AI provider’s context. The highest-risk action here is often retrieval itself.

A particularly useful test would remove a user from a private channel, resync, and prove that neither search nor cached thread collection can return the revoked content. That turns the local privacy claim into an enforceable lifecycle guarantee.