slack-mcp

A Model Context Protocol (MCP) server that enables AI assistants to interact with Slack workspaces. This server provides a bridge between AI tools and Slack, allowing you to read messages, post content, and manage Slack channels programmatically through MCP-compatible clients.

Disclaimer: This is not an official Red Hat tool.

What is this and why should I use it?

This MCP server transforms your Slack workspace into an AI-accessible environment. It provides 21+ tools for comprehensive Slack interaction:

Message & Thread Operations

• Read channel history with date filtering and thread support

• List unread direct messages with read markers (browser-session auth only; uses Slack's client.counts endpoint, which is not part of the public Web API)

• Post messages and replies to threads

• Search messages across workspace or within specific channels

• Execute Slack commands

Channel Management

• List, join, create, and rename channels

• Look up channel IDs by name

• Invite users to channels

Reactions & Users

• Add and view emoji reactions

• Send direct messages and group DMs

• Manage usergroups (clear members)

Utility

• Check authentication status

• Cache management for performance

Key Benefits

• Seamless Integration: Connect your AI assistant directly to Slack without manual copy-pasting

• Automated Workflows: Build AI-powered Slack bots that can read, analyze, and respond to messages

• Enhanced Productivity: Let AI help manage notifications, summarize conversations, or automate routine Slack tasks

• Real-time Collaboration: Enable AI assistants to participate in team discussions and provide instant insights

Use Cases

• Team Assistant: Have an AI that can read team updates and provide summaries

• Notification Manager: Automatically categorize and respond to incoming messages

• Knowledge Base: AI that can search through channel history and provide context

• Meeting Scheduler: AI that can read meeting requests and help coordinate schedules

Related MCP server: Slack MCP Server

Setting Up with Claude Code

This repo ships as a Claude Code plugin with a guided setup skill. Claude will walk you through the entire process — no manual config editing required.

Run the setup script. It handles everything — venv, Playwright, token extraction, wrapper script, and Claude Code registration. The only interaction required is logging in to Slack when the browser opens, and entering an optional channel ID for server logs, if desired.

python3 <(curl -fsSL https://raw.githubusercontent.com/redhat-community-ai-tools/slack-mcp/main/scripts/setup-slack-mcp.py)

Or clone the repo first and run it locally:

git clone https://github.com/redhat-community-ai-tools/slack-mcp
python3 slack-mcp/scripts/setup-slack-mcp.py

Options:

When tokens expire, just run:

python3 slack-mcp/scripts/setup-slack-mcp.py --refresh-tokens

Desktop App Token Refresh (Linux)

If you have the Slack desktop app installed, you can refresh tokens without opening a browser:

slack-mcp/scripts/slack-refresh-tokens --validate

This reads tokens directly from the desktop app's local storage on disk — no DevTools, no Playwright, no manual steps. Requires the Slack app to be signed in.

Requirements: python3, python3-cryptography, secret-tool (libsecret/gnome-keyring), curl, jq

This is useful for CI hooks or session startup scripts that need to silently refresh tokens before launching the MCP server.

Bot token authentication (recommended)

For better security, use a Slack App bot token (xoxb-) instead of browser session tokens. Bot tokens provide:

• Scoped access — only the OAuth permissions you grant, not full user access

• Distinct identity — actions appear as the bot, not as your user account

• Central management — IT can audit and revoke via the Slack admin panel

• No browser DevTools — tokens are generated once in the Slack App settings

Setup

1. Create a Slack App at api.slack.com/apps

2. Add OAuth scopes: channels:read, channels:history, channels:manage, groups:read, groups:history, groups:write, chat:write, reactions:read, reactions:write, search:read, users:read, commands, mpim:write

3. Install to your workspace and copy the Bot User OAuth Token (xoxb-...)

4. Invite the bot to channels it needs access to

Running with a bot token

Set SLACK_BOT_TOKEN instead of SLACK_XOXC_TOKEN/SLACK_XOXD_TOKEN:

{
  "mcpServers": {
    "slack": {
      "command": "podman",
      "args": [
        "run", "-i", "--rm",
        "-e", "SLACK_BOT_TOKEN",
        "-e", "LOGS_CHANNEL_ID",
        "quay.io/redhat-ai-tools/slack-mcp"
      ],
      "env": {
        "SLACK_BOT_TOKEN": "xoxb-...",
        "LOGS_CHANNEL_ID": "C7000000"
      }
    }
  }
}

LOGS_CHANNEL_ID is optional. When omitted, tool activity is written to stderr instead of posted to Slack.

If both SLACK_BOT_TOKEN and SLACK_XOXC_TOKEN/SLACK_XOXD_TOKEN are set, the bot token takes precedence.

Read-only mode

For agents or automation that should browse and search Slack without posting, reacting, running commands, or joining channels, enable read-only mode.

• Environment variable: set SLACK_MCP_READ_ONLY to a truthy value (1, true, yes, or on, case-insensitive).

• CLI: pass --read-only when starting the server (e.g. slack-mcp --read-only, equivalent to setting the variable).

In read-only mode, tools that mutate Slack state (post_message, send_dm, post_command, add_reaction, join_channel) raise a clear error. Read tools (history, search, threads, whoami, channel listing, cache refresh helpers, and so on) behave as usual. Tool activity that would normally be mirrored to LOGS_CHANNEL_ID is written to stderr instead.

On startup, the server logs a line to stderr when read-only mode is active.

For Podman or Docker, add -e SLACK_MCP_READ_ONLY=true (and the matching key in env) when you want the container to run read-only.

Running as a uv tool (local, no container)

Prefer not to run a container? Install slack-mcp as a uv tool. This puts a slack-mcp command on your PATH that runs the server directly.

Requires uv and Python ≥ 3.10.

# from a clone of this repo
uv tool install .

# or straight from git
uv tool install git+https://github.com/redhat-community-ai-tools/slack-mcp

Run slack-mcp --help for a summary of flags and environment variables.

Tokens from tokens.env (recommended)

slack-mcp reads Slack session tokens from ~/.local/share/slack-mcp/tokens.env when they are not already set in the environment — the same file the token tooling writes:

scripts/slack-refresh-tokens          # or: python3 scripts/setup-slack-mcp.py

Then the MCP client config needs no env block — just the command:

{
  "mcpServers": {
    "slack": {
      "command": "slack-mcp"
    }
  }
}

The file uses SLACK_MCP_XOXC_TOKEN / SLACK_MCP_XOXD_TOKEN; the server maps those onto the SLACK_XOXC_TOKEN / SLACK_XOXD_TOKEN it uses. Override the path with SLACK_MCP_TOKENS_FILE. Real environment variables always take precedence over the file, so you can still override per-client.

Tokens from the client config (alternative)

To keep tokens in the MCP config instead of a file, pass them in env:

{
  "mcpServers": {
    "slack": {
      "command": "slack-mcp",
      "env": {
        "SLACK_XOXC_TOKEN": "xoxc-...",
        "SLACK_XOXD_TOKEN": "xoxd-...",
        "LOGS_CHANNEL_ID": "C7000000"
      }
    }
  }
}

A bot token works the same way — set SLACK_BOT_TOKEN in env instead of the xoxc/xoxd pair. If the client does not inherit your PATH, use the absolute path (~/.local/bin/slack-mcp after uv tool install), or launch via uv run --directory /path/to/slack-mcp slack-mcp.

The user cache and the default tokens.env live under ~/.local/share/slack-mcp/ (override with SLACK_MCP_DATA).

Claude Code (bot token):

claude mcp add slack -e SLACK_BOT_TOKEN=xoxb-... -- slack-mcp

Or, with tokens.env already in place, no env needed:

claude mcp add slack -- slack-mcp

Running with Podman or Docker

You can run the slack-mcp server in a container using Podman or Docker:

Example configuration for running with Podman:

{
  "mcpServers": {
    "slack": {
      "command": "podman",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e", "SLACK_XOXC_TOKEN",
        "-e", "SLACK_XOXD_TOKEN",
        "-e", "MCP_TRANSPORT",
        "-e", "LOGS_CHANNEL_ID",
        "quay.io/redhat-ai-tools/slack-mcp"
      ],
      "env": {
        "SLACK_XOXC_TOKEN": "xoxc-...",
        "SLACK_XOXD_TOKEN": "xoxd-...",
        "MCP_TRANSPORT": "stdio",
        "LOGS_CHANNEL_ID": "C7000000"
      }
    }
  }
}

LOGS_CHANNEL_ID is optional. When omitted, tool activity is written to stderr instead of posted to Slack.

Activity logging

By default, tool activity is written to stderr (visible in your terminal or process logs). To mirror activity to a Slack channel instead, set LOGS_CHANNEL_ID to any channel the bot or session user has access to — a self-DM or a DM with Slackbot works well for personal use.

LOGS_CHANNEL_ID=C7000000

In read-only mode, LOGS_CHANNEL_ID is ignored and all activity is always written to stderr.

Running with non-stdio transport

To run the server with a non-stdio transport (such as SSE), set the MCP_TRANSPORT environment variable to a value other than stdio (e.g., sse).

Example configuration to connect to a non-stdio MCP server:

{
  "mcpServers": {
    "slack": {
      "url": "https://slack-mcp.example.com/sse",
      "headers": {
        "X-Slack-Web-Token": "xoxc-...",
        "X-Slack-Cookie-Token": "xoxd-..."
      }
    }
  }
}

Extract your Slack XOXC and XOXD tokens easily using browser extensions or Selenium automation: https://github.com/maorfr/slack-token-extractor.