Slackの過去ログをClaude DesktopやCursorから検索できるようにする（MCP・ローカル完結）This article explains how to back up Slack conversations locally using the…

Locating a specific past conversation in Slack can be surprisingly difficult, particularly when the message you need sits somewhere in months of channels and threads. This article describes a setup that lets you pull those old exchanges directly from AI clients such as Claude Desktop or Cursor, using the Model Context Protocol (MCP) while keeping all of the data on your own computer.

The described workflow has three stages. First, Slack conversations are backed up locally to your PC with a Windows application called Empowia for Slack. Second, that local archive is exposed as an MCP server. Third, any MCP-compatible client—Claude Desktop and Cursor are the examples cited—connects to that server and can search and reference the archived messages in read-only mode. Because nothing is uploaded to the cloud, the author frames the approach as local-first, and it notably does not require Slack administrator privileges or the issuance of an API token.

The absence of admin rights and tokens is a meaningful distinction. Many existing Slack integrations rely on the Slack Web API, which typically means creating an app, obtaining OAuth scopes, and having a workspace administrator approve access. That path can be a non-starter in organizations with strict governance. By backing up conversations to the local machine and serving them from there, this method sidesteps those steps, though it also means the archive reflects only what the user can already see and export rather than a full workspace dataset.

MCP is the connective layer that makes this possible. Introduced by Anthropic in late 2024 and since adopted by a range of tools, it is an open standard for connecting AI assistants to external data sources and capabilities. An MCP server advertises resources and tools that a compatible client can call, which lets the model reason over content it would not otherwise have in its context window. Anthropic has likened MCP to a universal connector for AI applications, and support now appears in clients beyond Claude Desktop, including Cursor and several other editors and assistants. In this configuration the Slack archive is simply another data source that a client can query on demand.

Running the server in read-only mode is a sensible safety choice. It means the AI client can search and cite messages but cannot alter or delete the underlying archive, reducing the risk of accidental data loss when an assistant is given broad instructions. Keeping the archive local also limits exposure, since message content is not transmitted to a third-party service for indexing, which may matter for teams handling sensitive internal discussions. Users should still bear in mind that whatever client they connect may send query results to a model provider as part of normal operation, depending on how that client is configured.

Some context on the surrounding ecosystem helps. Slack offers its own export tools, but comprehensive exports are generally gated behind administrator access and plan level, and the company has also been layering AI-powered search into the product itself. Separately, community-built MCP servers for Slack already exist, most of which authenticate through the Slack API and therefore inherit the token and permission requirements this workflow aims to avoid. The Empowia approach trades that live connectivity for a self-contained local copy, which favors privacy and independence at the cost of not reflecting real-time changes in the workspace.

The tooling here is Windows-specific, so macOS and Linux users would likely need an alternative for the backup step. As always with locally stored communication history, the resulting archive should be protected like any other sensitive file, since it consolidates potentially confidential messages in one place on disk.

One disclosure noted in the source is that Empowia for Slack is connected to the article's author, so readers may wish to weigh that when evaluating the recommendation. For individuals or small teams who want to make their Slack history searchable through an AI assistant without touching cloud services or administrative controls, the described local-first, MCP-based configuration offers a concrete and relatively lightweight path.