Slack Claude Connector: The Complete Guide

Slack MCP is a secure integration that connects your Slack workspace to AI clients including Claude and Cursor, giving language models direct access to your real Slack data through a defined set of supported actions. The connector is documented on Slack's developer site with a dedicated "Connect to Claude" guide, and there is a rich ecosystem of tutorials covering both Claude Desktop and Claude Code.

The connector supports the day-to-day Slack actions you'd want an agent to take:

• Read messages from channels or threads.
• Post messages to any channel or reply within threads.
• Search messages by keyword, timestamp or sender.
• List users or channels and fetch user info to personalise content.
• Automate updates - send messages, create reminders, and keep the workspace organised.

• Catch-up summaries: "what happened in #general while I was in meetings?" - Claude summarises the thread instead of you scrolling.
• Cross-channel search: find every conversation about a topic across all channels and produce a briefing.
• Meeting prep: "What did the product team decide about feature priority last week?" - Claude searches and answers.
• Lead and ticket notifications: when a new lead lands in your CRM, Claude posts a formatted summary to your sales channel automatically.
• Thread follow-ups: summarise a long thread and post the summary plus next steps back into the same thread.

The single most cited Slack-connector workflow - turning a backlog of unread channels into a five-minute briefing instead of an hour of scrolling:

1. Monday morning, hundreds of unread messages. You ask: "Summarise what happened in #product, #engineering and #incidents since Friday."
2. Claude reads the channels and produces a per-channel digest, pulling out decisions, blockers and anything addressed to you.
3. You ask a follow-up: "What did the team decide about the pricing experiment?" - Claude searches across channels and quotes the relevant thread.
4. You spot an action: "Reply in the pricing thread that I'll own the rollout and post a summary of next steps." - Claude drafts and posts it back into the same thread.
5. For the parts you missed in #incidents, you ask Claude to list the open follow-ups so nothing falls through.

Read, search, synthesise, respond - all without leaving the conversation. For anyone whose Slack is the real source of truth for what the team is doing, this loop alone justifies the connector.

Reception has been strongly positive among teams that use Slack as their primary nervous system. The two most cited wins are catch-up summaries (no more scrolling 200 messages) and cross-channel search (finding every conversation on a topic). Setup guides from Composio, Strac, Digidog and Slack's own docs make the path well-trodden.

The trade-off, candidly noted in community write-ups: a write-capable Slack agent in a busy workspace needs careful scoping, because an accidental broad post is visible immediately. Start with read access and a couple of safe channels for posting.

Beyond reading and posting, Anthropic supports Claude Code in Slack - so engineering teams can run the agent inside the same channels where they already discuss work. That extends the connector from "summarise and respond" to "do the work" - kicking off Claude Code tasks, reviewing diffs and shipping changes, all in the channel.

Slack runs an official hosted MCP server at https://mcp.slack.com/mcp. It uses confidential OAuth, so connecting is a consent flow rather than pasting a token, and the agent acts within your Slack permissions.

Claude Desktop and claude.ai (easiest)

On the Customize page, click Connectors, then + to add a connector. Find Slack, add it, and authenticate to your workspace via OAuth. The Slack tools appear once consent completes - no config file needed.

Claude Code

Add the remote server and authenticate with /mcp:

claude mcp add --transport http slack https://mcp.slack.com/mcp

The official Slack plugin (the slackapi/slack-mcp-plugin repo) ships a ready-made .mcp.json with the server URL, client ID and a fixed callback port - useful if you want a checked-in team config rather than per-developer setup. Whichever route you take, grant only the scopes you need and pilot in a safe channel before enabling write across the workspace.

Slack's existing permission model composes with the connector: the agent can only see and post to what your account can. Within that, scope tightly. A misfired write is publicly visible; a misfired read isn't great either if it surfaces sensitive content. Read-first, write-by-channel, and review the action history are the safe defaults.

Slack's connector is widely deployed and well-documented, but a few issues recur - most of them about workspace admin approval and scopes rather than the connector itself.

"App not approved for this workspace"

Many Slack workspaces restrict which apps members can install. If the connector won't authorise, a Slack workspace admin needs to approve the Claude app in the workspace's app-management settings. In tightly governed orgs this is the single most common blocker - raise a request with your Slack admin rather than retrying the OAuth flow.

Search returns nothing

Slack search through the MCP only covers channels the authenticated user is a member of - the agent cannot search channels you haven't joined, and private channels you're not in are invisible. If a "find every conversation about X" query comes back thin, join the relevant channels first, or name a specific channel in the prompt.

An accidental broad post goes out

Unlike most connectors, Slack writes are instantly public. A loose "post an update" prompt can land in the wrong channel in front of the whole company. Start read-only, enable posting on a small set of safe channels, and ask Claude to confirm the target channel before it sends anything during early use.

Rate limits on large workspaces

Broad searches and bulk reads across a big workspace can hit Slack's API rate limits, which surface as slow or partial responses. Scope queries by channel and time window rather than asking the agent to sweep everything, and break large catch-up jobs into smaller asks.

Missing message history

On free and lower-tier Slack plans, message history is capped, so the agent simply cannot read beyond the retained window - older threads return nothing. This is a Slack plan limitation, not a connector fault; upgrading the workspace plan is the only fix.

The Slack MCP server is provided by Slack at no additional charge - connecting Claude does not add a Slack licence fee, and the agent's reads and posts count against your workspace exactly as your own activity would. What is gated is the underlying Slack plan: searchable message history depth, some search features and admin app-approval controls all vary by tier (Free, Pro, Business+, Enterprise Grid).

The connector is generally available and hosted by Slack, so there is nothing to run or update. On Enterprise Grid, expect admin approval and org-wide app policies to govern whether and how the connector can be installed - plan the rollout with your Slack admins rather than enabling it ad hoc.

• Public surface: Slack actions are visible in real time, so write errors are visible errors.
• Plan-dependent features: some workspace features (e.g. message history depth) depend on your Slack plan.
• Volume: very large workspaces may hit rate limits on broad searches; scope queries.

Any team where Slack is the nervous system - which is most knowledge-work organisations today. Engineering teams in particular get an extra dimension via Claude Code in Slack.

What can it do?

Read, post, reply, search and list across your Slack workspace, and fetch user info to personalise responses.

Can I run Claude Code in Slack?

Yes - Anthropic supports running Claude Code inside Slack channels.

Is it safe?

Permissions compose with Slack's own model; scope tightly, start with read, and review actions.

Who is it for?

Any Slack-first team - engineering, product, support, sales.

The Slack connector turns Claude from an outside assistant into a participant in your team's daily conversations - summarising, searching, posting and, with Claude Code, even shipping work. For Slack-first teams it is among the highest-leverage connectors available.

Start in read mode, expand carefully, and let Claude work where your team actually talks. More in the complete Claude Connectors Directory.