AI Agents Guide

AI Agent for Slack: What It Can Do and How to Set One Up

Slack now sells its own AI agent, and Anthropic and OpenAI offer theirs. Here is what each can do, what it costs and how to switch one on without over-sharing.

Flat illustration of an AI agent for Slack: a chat window linked by dotted lines to a green agent hub, a calendar, a checklist, a document and a shield with a tick.

AI agent for Slack: the short answer

An AI agent for Slack is AI that can read your workspace conversations and act on them, such as answering questions, summarising threads, posting updates or changing records in other apps. Slack also builds its own agent, Slackbot, into paid plans. You can also connect Claude, ChatGPT and other tools through official apps or the Slack MCP server, which admins must approve. If your team keeps its docs in Notion, our guide to an AI agent for Notion covers the same questions for that app.

Choosing an AI agent for Slack no longer means hunting for a clever bot in the app directory. Slack itself now sells an agent, while the big AI companies also ship their own Slack apps. As a result, a small team can easily end up with three or four assistants reading the same channels.

That matters because Slack holds a company’s informal memory: decisions, customer complaints, half-finished plans and the occasional file that should never have been shared. So the useful question is not only what an agent can do, but also what it can see and who approved it.

This guide explains the main options as of October 2026, what each one can realistically do, what it costs where prices are public, and how to set one up without handing over more access than it needs. We drew on documentation from Slack, Anthropic and OpenAI, so TheJusGrow has not tested these products for this article.

What counts as an AI agent for Slack?

In plain terms, an agent is AI that can take steps on its own instead of only answering a question. In Slack, that usually means three abilities: it can search messages and files, it can write or post in channels, and it can reach other tools such as a CRM, an issue tracker or a calendar.

A simple bot that replies when you mention it sits at one end of that range. However, at the other end are agents that join channels, keep memory between conversations and start work without being asked. Most products sit somewhere in between, so check four things before you switch one on:

  • Reading access: public channels only, or private channels and direct messages too.
  • Actions: drafting only, or posting, scheduling and updating records directly.
  • Identity: your own permissions, or a separate service account.
  • Billing: a per-seat plan, a credit balance or a usage meter.

For technical background, our guide to the Model Context Protocol explains the open standard that many of these connections now use.

Slackbot: the AI agent built into Slack

Salesforce announced the general availability of the new Slackbot on January 13, 2026, for Business+ and Enterprise+ customers. At the time, it described Slackbot as a personal work agent that uses your existing conversations, files and permissions. At launch, the company listed jobs such as finding information, drafting meeting notes and project updates, scheduling meetings and setting reminders.

Since then, Slack’s product page has gone further. It says Slackbot can research across Slack history, connected apps and the web, build repeatable workflows from a conversation, and take action in integrated apps such as Jira, Figma, Salesforce and DocuSign.

Then, at Dreamforce in September 2026, Slack added more. According to its announcement, Slackbot runs on Claude, has passed one million monthly active users, and now has a full-screen “Big mode” for longer research and writing. Slack also said people can tag Slackbot into a channel so the whole team builds on one answer, although that channel feature was still in an early pilot.

On privacy, Slack’s Slackbot page says the agent inherits your organisation’s permissions and access controls, and it states that your data is never used to train AI models. Those are company claims, so admins should still read their own contract and settings.

Slack planSlackbot access (Slack’s description)Listed US price per user per month
FreeLimited trial$0
ProLimited trial$7.25 billed annually, or $8.75 monthly
Business+Ongoing access with weekly message limits$15 billed annually, or $18 monthly
Enterprise+Unlimited accessContact sales

These prices come from Slack’s US pricing page in October 2026, so they may differ by country. When we checked, Slack was also advertising a temporary discount on monthly billing, so confirm the current figure before you budget.

Can you use Claude or ChatGPT as an AI agent for Slack?

Yes, you can. In fact, if your team already pays for another assistant, you can usually bring it into Slack rather than switching. For now, the two best-documented options come from Anthropic and OpenAI.

Claude Tag

Anthropic’s Slack product now goes by the name Claude Tag. Its help centre says the earlier Claude in Slack app switched over to Claude Tag on August 3, 2026. According to Anthropic’s setup guide, Claude Tag is in public beta and works only on Claude Team and Enterprise plans, not on Free, Pro or Max.

Setup needs two admins, because each side controls its own system. First, a Slack workspace admin installs the app and runs a pairing command. Then a Claude organisation owner redeems the code, picks channels and sets a monthly spend limit, because channel work draws from the organisation’s usage balance rather than individual seats.

Anthropic also recommends giving Claude its own accounts in other tools, with the narrowest role that covers the work. That way, its actions show up in each tool’s audit log under its own name, and you can cut off its access without touching anyone else’s.

@ChatGPT

OpenAI’s help centre describes @ChatGPT for Slack and Microsoft Teams as an organisational assistant for Business, Edu and Enterprise accounts. It runs under its own service account instead of inheriting each user’s permissions. Admins then choose which channels and company tools it can use, and they can also add optional spend caps.

However, one detail is easy to miss. OpenAI notes that channel restrictions apply to Slack channels but do not restrict direct messages. Therefore, if you limit @ChatGPT to a few channels, check separately what people can ask it in DMs. For OpenAI’s wider workplace plans, also see our ChatGPT Work explainer.

Other marketplace agents

Slack says more than 50 partners build agents on its platform, including Google, OpenAI and Perplexity. Still, we have not reviewed them individually here. Instead, treat each feature list as a company claim and read the permissions screen before installing anything.

How the Slack MCP server connects outside agents

Meanwhile, the Slack MCP server is the plumbing behind many of these connections. Slack made it generally available, together with its Real-Time Search API, in February 2026. In effect, it gives an outside AI agent for Slack a standard way to search and act in Slack using the signed-in person’s permissions.

For example, according to Slack’s developer documentation, the server’s tools can:

  • search messages, files, users and channels;
  • read full channel histories and threads;
  • send, schedule and draft messages;
  • create and update canvases and lists;
  • upload files and add emoji reactions.

Slack lists Claude.ai, Claude Code, Perplexity and Cursor as ready-made clients, so you do not need to write code to use it from those apps. However, only apps published in the Slack Marketplace or built internally may use it, and admins approve MCP clients through the normal app approval process.

So that approval step is the main safeguard. Because the server can send messages and read entire channel histories, an agent connected this way can do almost anything you can do in Slack. Still, Slack says it respects user permissions, so it cannot open channels you cannot open.

What can an AI agent for Slack do for a small team?

Overall, the documented features point to a handful of practical jobs for an AI agent for Slack. None of these needs a developer, although several need an admin to approve an app first.

TaskWhat the agent doesOptions that document it
Catching upSummarises a busy channel or threadSlackbot, Claude Tag, @ChatGPT
Finding decisionsSearches past messages and links the sourcesSlackbot, MCP clients such as Claude and Perplexity
Status updatesDrafts a project update from channel activitySlackbot, Claude Tag
Updating other appsChanges a record in a connected tool after a conversationSlackbot with integrated apps, Claude Tag with tool access
Shared listsCreates or updates a Slack list of tasksSlack MCP server clients

A worked example shows how an AI agent for Slack fits together with a real team. For instance, imagine a five-person online shop that runs customer support in a #support channel. First, the owner asks the agent each Monday to list unresolved complaints from the past week, with links. Next, the agent drafts a short summary for the team channel, and a person checks it before posting. Finally, the agent builds a Slack list of the three most common issues so the team can assign owners.

Every step there uses documented features, because nothing here needs custom code. Even so, the human checks are the point: the agent gathers and drafts, while people decide what to tell customers. Also, if those customer records live in a CRM, our guide to an AI agent for HubSpot covers that side.

How the main options compare

OptionMade byPlan neededHow it is billed
SlackbotSlackBusiness+ for ongoing use; Enterprise+ for unlimited useIncluded in the Slack plan, with weekly limits on Business+
Claude TagAnthropicClaude Team or Enterprise (public beta)Organisation usage credits, with a monthly spend limit
@ChatGPTOpenAIChatGPT Business, Edu or EnterpriseSet by admins, with optional spend caps
Slack MCP serverSlack, used by Claude, Perplexity, Cursor and othersDepends on the AI app, plus Slack approvalDepends on the AI app

In practice, the right AI agent for Slack mostly depends on what you already pay for. A team on Business+ can start with Slackbot at no extra cost, while a team already standardised on Claude or ChatGPT may prefer to keep one assistant everywhere.

How to set up an AI agent for Slack safely

The setup steps differ by product, but a safe order is the same. So here is a starter checklist based on the vendors’ own documentation:

  1. Pick one job. Weekly channel summaries are a better start than “help with everything”.
  2. Check your plan. Slackbot needs Business+ for ongoing use, Claude Tag needs Claude Team or Enterprise, and @ChatGPT needs ChatGPT Business, Edu or Enterprise.
  3. Begin read-only. OpenAI advises starting with read access when that is enough and adding write access only when needed.
  4. Limit channels. Add the agent to one pilot channel and keep it out of HR, finance and legal channels at first.
  5. Cap spending. Both Claude Tag and @ChatGPT let admins set spending limits.
  6. Watch the logs. Slack offers audit logs for MCP activity, and admins can remove approved apps.
  7. Agree a house rule. For example: the agent drafts, and a named person sends.

It is also worth clearing out old integrations while you are there. Similarly, our guide on how to review apps connected to your accounts walks through the same idea for personal accounts.

What are the risks of giving an agent access to Slack?

Slack is a high-value target because so much sits in it. As a result, the main risks fall into four groups.

First, over-sharing. An agent that can read private channels and DMs can surface something to the wrong person if access is set loosely. Slack says its MCP server respects user permissions, but a shared service account, as with @ChatGPT, follows whatever access its admin grants.

Second, prompt injection. Agents also read text written by other people, including external guests in shared channels. A message could hide instructions aimed at the agent, so approval steps before it sends, posts or spends really matter. Our prompt injection explainer shows how these attacks work and why they are hard to block.

Third, mixing tools. Slack’s developer guide warns users to think carefully before combining its MCP server with other MCP servers, since each has its own security characteristics.

Fourth, cost creep. Credit-based agents can run up bills quietly, so set alerts and caps from day one.

AI agent for Slack: who it suits and who should wait

An AI agent for Slack makes most sense for teams that already live in Slack and lose time catching up. For example, it suits remote teams across time zones, support teams answering repeat questions, and managers who write weekly updates from channel chatter.

On the other hand, you may want to wait if your workspace mixes clients or external guests in shared channels. The same goes for teams that handle regulated data without clear contract terms, or Free and Pro users who only need the odd summary. In those cases, a free trial or a tightly limited pilot is the sensible first step.

Key takeaways

  • Slackbot is Slack’s own AI agent; ongoing use needs Business+ or Enterprise+, while Free and Pro get a limited trial.
  • Claude Tag and @ChatGPT bring outside assistants into Slack, but both need business plans and admin setup.
  • The Slack MCP server lets approved AI apps search and act in Slack with the user’s own permissions.
  • Start with one task, one channel and read-only access, then add write access and spend caps deliberately.
  • Treat vendor privacy statements as company claims and check your own contract.

AI agent for Slack: FAQs

Is Slackbot AI free?

Not fully, because Slack limits it by plan. Slack says Free and Pro plans get a limited trial of Slackbot, Business+ gets ongoing access with weekly message limits, and Enterprise+ gets unlimited access, as of October 2026.

Can ChatGPT read my Slack messages?

Only when your workspace connects it. OpenAI says @ChatGPT for Slack works on Business, Edu and Enterprise accounts and runs under a service account whose access an admin controls.

What is the Slack MCP server?

It is Slack’s official server, and it lets approved AI apps search Slack, read and send messages, and manage canvases and lists, using a person’s own Slack permissions. Slack made it generally available in February 2026.

Does Slack train AI on my messages?

No, according to Slack. Its Slackbot page says your data is never used to train AI models. That is the company’s statement, so check your plan’s terms for the details.

Sources