Claude in Slack
A beginner-friendly guide to working with Claude inside Slack: the earlier Claude app, the newer Claude Tag teammate, how each one works, what you can hand over, how access is controlled, and how to use it safely as a team.
Imagine a new colleague who sits in the same chat room as the rest of your team. They read the thread when you ask them to, they answer questions, they draft the summary nobody had time to write, and when a bigger job comes along, they take it away, work on it and report back in the same conversation, so everyone can see what happened. Nobody has to open a different tool or copy and paste anything between windows.
That is the idea behind Claude in Slack. Slack is where many teams already talk, decide and coordinate, so bringing Claude into it means help arrives where the work already is. This guide explains it from scratch. No technical knowledge is needed. Every new word is explained when it first appears, and each big idea comes with an everyday comparison.
This area changed quickly in 2026. An earlier Claude app for Slack was replaced by a newer product called Claude Tag, and admin settings, plans and features are still evolving. The details here reflect Anthropic’s help pages and press coverage at the time of writing. Before you install or rely on anything, check the current Claude Tag documentation on claude.com and your Slack workspace’s own settings.
1What Does “Claude in Slack” Mean?
The phrase can point to more than one thing, so let us separate them first.
Slack is a team messaging tool where people talk in channels, which are shared rooms organised by topic or project, and in threads, which are side conversations attached to a single message. It also has direct messages, often called DMs, for private one-to-one chats. Many companies use Slack as the main place where work gets discussed.
There are three ways Claude and Slack connect. First, there is the earlier Claude app for Slack, which let individual users chat with Claude from inside Slack. Second, there is Claude Tag, the newer product that makes Claude a shared teammate in selected channels. Third, there is the Slack connector, which works in the opposite direction: it lets Claude, in its own apps, search and refer to your Slack messages when you ask.
Earlier Claude app
A per-user app for DMs, an assistant panel and @mentions. Replaced by Claude Tag from August 3, 2026.
Claude Tag
The current product: one Claude identity, set up by an admin, that the whole team can tag into work.
Slack connector
Lets Claude in its own apps search and reference Slack messages, with your permissions.
Claude Code in Slack
Routes coding tasks from Slack to Claude Code, for teams with that access.
Think of an office. In the first version, each employee had a private assistant they could whisper to. In the newer version, there is one shared team member whom anyone can ask for help in the open. And the connector is like letting your own assistant read the office noticeboard when you ask.
When people say “Claude in Slack” today, they usually mean Claude Tag. When you read older articles, they often describe the earlier app, so check the date.
2A Short Timeline
Understanding what came when clears up most of the confusion.
Anthropic first released a Claude app for Slack in beta, describing it as able to summarise threads, answer questions and more. In that early design, Claude only saw messages in threads where it had been explicitly mentioned, or in DMs sent to it. Later, Anthropic introduced two ways to combine Claude with Slack: adding Claude directly to a workspace, and connecting Slack to the Claude apps. Coverage dates that announcement to October 2025.
In June 2026, according to press coverage, Anthropic launched Claude Tag, describing it as a new way for teams to work with Claude, beginning with Slack. Anthropic’s help pages say that Claude in Slack would be switched over to the new Claude Tag experience on August 3, 2026, and that people who want to integrate Claude and Slack should use Claude Tag. Coverage also says administrators could opt in to migrate within a limited window.
Earlier Claude app for Slack
Released in beta, letting individuals chat with Claude, summarise threads and ask questions inside Slack.
Two-way integration, around October 2025
Claude could be added to Slack, and the Slack connector let Claude’s own apps reference Slack content.
Claude Tag launch, June 2026
Reported as a persistent, proactive, multiplayer teammate, starting with Slack.
Switchover, August 3, 2026
Per Anthropic’s help pages, the earlier app was replaced by the Claude Tag experience.
flowchart LR
A["Earlier Claude app: per-user helper"] --> B["Slack connector: Claude reads Slack on request"]
B --> C["Claude Tag: shared teammate in channels"]
C --> D["Earlier app replaced from August 3, 2026"]
Many shops start with a small counter service, then add online ordering and later redesign the whole store. The name stays familiar, but the experience behind it changes. Claude in Slack has gone through a similar redesign.
3The Earlier App: Claude as a Personal Helper
Because many guides still describe the first version, it is worth understanding how it worked.
The earlier Claude app for Slack was available to users on paid Slack plans, and Slack administrators had to approve the app before individuals could use it. Each person connected their own Claude account. There were several ways to interact. You could send Claude a direct message for private, one-to-one help. You could open an assistant panel on the side of Slack. And you could mention Claude by name in a channel or thread to ask a question there.
Anthropic’s pages described a set of capabilities similar to the Claude web app, such as web search, document analysis and use of tools you had already connected. When you asked Claude for help in a public thread, it was described as drafting responses privately first, so you could review, edit or regenerate before sharing with your team. Some features of the web app were not available in that app, such as Research and extended thinking.
Direct messages
Private chats with Claude, useful for drafting and thinking before sharing anything.
Assistant panel
A side panel in Slack where you could chat with Claude while viewing your channels.
Mentions in channels
Mentioning Claude in a thread so it could respond to a question or summarise discussion.
Private drafts first
In public threads, responses were drafted privately so you stayed in control of what was shared.
The earlier app was like having your own personal notepad assistant. It helped you write and think, and you decided what to share with the group. It served the individual more than the team.
If you read guides that describe the assistant panel and per-user setup, remember that they describe the earlier app. For the current experience, look for Claude Tag.
4Claude Tag: Claude as a Teammate
Claude Tag changes the picture from a personal helper to a shared team member.
Anthropic’s support materials describe Claude Tag as giving your team one Claude identity, set up by an administrator. Press coverage describes it as a way to tag @Claude into work in selected Slack channels. Claude then breaks the task into stages, works with the tools and data it has been given access to, and posts the results in the thread, often tagging you back when it is done. Anthropic describes it as an evolution of Claude Code that is more proactive and built to work with a full team, and its product lead was quoted calling it Anthropic’s first product that is natively multi-player and proactive.
The key idea is multiplayer. In a personal chat, only you see the conversation. In Claude Tag, Claude works in the open in a channel, so teammates can see the request, follow the progress, add details or correct course. Because the work happens in the channel, the knowledge stays with the team.
| Aspect | Earlier app, personal style | Claude Tag, team style |
|---|---|---|
| Who it serves | The individual user | The whole team in a channel |
| Identity | Your own Claude account | One Claude identity set up by an admin |
| Visibility | Mostly private, with drafts shared by you | Work visible in the channel thread |
| Style of work | Responds when you ask | Delegated tasks, plus proactive follow-ups |
| Memory | Per user and session | Described as persisting in the channel across days |
A shared whiteboard in a project room is different from a private notebook. Everyone can see what is written, add to it and build on it. Claude Tag works like a team member standing at the whiteboard, not someone whispering in one person’s ear.
Anthropic has said it plans to extend Claude Tag beyond Slack to other collaboration platforms. For now, Slack is where it is available.
5How Claude Tag Works Behind the Scenes
A simple picture helps you understand why access settings matter so much.
When someone tags Claude in a channel, Claude reads the relevant context, plans the work, carries it out in a private working environment, and posts progress and results back in the thread. Press descriptions say the work runs in an isolated sandbox on Anthropic’s infrastructure, and that Claude reaches outside tools, such as code repositories or data sources, only through what the administrators have allowed.
Coverage also describes two behaviours that make it different from an ordinary chatbot. The first is persistent channel memory, meaning Claude can carry context in a channel from one day to the next. The second is a proactive or ambient style, where Claude can follow up, monitor things and come back to you without being asked each time. How far these behaviours go depends on how your administrators configure them.
flowchart TD
A["Teammate tags @Claude in a channel"] --> B["Claude reads the thread and allowed context"]
B --> C["Plans the task in stages"]
C --> D["Works in an isolated environment using allowed tools"]
D --> E["Posts progress and results in the thread"]
E --> F["Tags the requester back when done"]
B -.-> G["Channel memory carries context across days"]
Tag and describe the task
Mention Claude and say what you want, with a link or the thread for context.
Claude plans and works
It breaks the job into steps and uses the tools that administrators have allowed.
Progress appears in the thread
Teammates can see what is being done and chip in with corrections or details.
You review the result
Claude posts the outcome and tags you. You check it, ask for changes or accept it.
Think of handing a job to a colleague at the start of the day. They read the brief, go off to work, send you short updates and then come back with the finished piece. You are free to do other things in the meantime. Delegating to Claude Tag follows the same rhythm.
6What You Can Hand Over
The range of tasks depends on what your workspace has connected and allowed.
Catching up on a long thread
Ask Claude to summarise a busy discussion, list the decisions made and note what is still open, so latecomers can catch up quickly.
Answering questions from shared knowledge
Ask Claude to find the relevant discussion or document, and answer a question based on what your team has shared.
Drafting content
Ask for a first draft of an update, announcement, reply or summary based on the discussion, which people can then edit.
Preparing for meetings
Ask Claude to gather updates and action items from relevant conversations and prepare talking points.
Onboarding new teammates
Ask Claude to help a newcomer understand a project by reviewing channel history and shared documents.
Data and analysis requests
Where a data source has been connected, ask Claude to pull figures and post them in the thread. Press coverage gave examples of such requests.
Technical work
For teams that have connected code tools, press coverage describes Claude diagnosing a problem and opening a draft change for review straight from a thread.
What Claude can do in your workspace depends on the channels, tools and data your administrators have connected. If it cannot do something, the setting may simply not be enabled.
A new colleague can only use the systems they have been given access to. A very capable person with no login can still help with thinking and writing, but cannot reach the database. The same is true here.
7Writing Good Requests in Slack
Clear requests work even better in a shared channel, where others will read them too.
State the goal
Say what you want to end up with, and why it matters.
Point to context
Link the specific thread, message or document Claude should use. The help pages suggest providing a link to steer Claude to a particular message or resource.
Describe the result
Say the format and length you expect, such as five bullet points or a short update.
Give limits
Mention what Claude should not touch, and anything that needs approval first.
Say who it is for
Tell Claude the audience, such as the leadership team or a new colleague.
Keep it to one job
Split big requests into separate threads so progress stays easy to follow.
Vague request
“@Claude can you look at this?”
Clear request
“@Claude please summarise the discussion in this thread for the leadership team. List the decisions made, the open questions and who owns each next step, in no more than six bullet points.”
Handing a colleague a sticky note that says “look at this” invites guesswork. Handing them a short brief with the goal, the file and the deadline lets them start straight away. The same applies to tagging Claude.
8Setting It Up
Setup is mainly an administrator’s job, and it is worth understanding in outline.
Claude in Slack has always required the Claude app to be approved by Slack workspace administrators. Claude Tag goes further: coverage says it is configured by an owner in Claude’s admin settings, as part of a short setup that provisions the identity Claude will use, and that installing the Slack app is a prerequisite. Because details of this flow may change, the official Claude Tag documentation is the right source for exact steps.
Slack admin approves the app
A Slack administrator installs or approves the Claude app for the organisation or chosen workspaces.
Owner configures Claude Tag
A Claude owner sets up Claude Tag in the admin settings and provisions the identity it will use.
Choose channels and tools
Admins decide which channels Claude may work in, and which tools and data it can reach there.
Set limits and oversight
Coverage describes controls such as spend limits and audit logs. Confirm which are available to you.
Teammates start tagging
Once configured, members can tag Claude in the allowed channels.
If your Slack requires approval and you are not an administrator, you may see a request to install prompt. Contact your Slack admin to approve the app and ask who owns the Claude setup.
Before a new colleague starts, the company sets up their email, gives them a badge and decides which rooms they can enter. Setting up Claude Tag is the same onboarding, done once by someone with the authority to do it.
9Permissions and Identity: Who Is Claude Acting As?
The most important design question for a shared agent is whose permissions it uses.
According to coverage of Claude Tag, Claude does not simply act as whichever user tagged it. In shared channels it uses an identity provisioned by administrators, such as a Claude Slack identity and service accounts for tools. Administrators scope what that identity can reach, channel by channel. Reports describe a model in which broad, low-risk tools can be available to the shared channel identity, while more sensitive and personal tools stay in private DMs under your own identity.
The reason is protection. If a shared agent used everyone’s personal access, anyone in a channel might be able to reach someone else’s private files through it. Giving the agent its own limited identity means it can see only what it has been given, and nothing more.
flowchart TD
A["Channel request"] --> B["Claude's own admin-provisioned identity"]
B --> C["Only the tools and data allowed for this channel"]
D["Private DM"] --> E["Your own identity and personal tools"]
C --> F["Shared work visible to the team"]
E --> G["Personal work visible to you"]
Why a separate identity helps
- Limits what the shared agent can reach
- Stops personal files leaking through a channel
- Makes it clear what Claude has access to
- Lets admins adjust access channel by channel
Things to keep in mind
- Claude may lack access to something you personally can use
- Admins must set the scope thoughtfully
- Too much access creates risk, and too little limits usefulness
- Details can vary by configuration and may change
A shared team assistant gets a team badge that opens only the team’s rooms, not the master key of whoever happens to ask. That way, a request from one person cannot open another person’s locked drawer.
10Claude Code in Slack and the Slack Connector
Two related features are often mentioned alongside Claude Tag, so here is how they differ.
Claude Code in Slack
Claude Code is Anthropic’s tool for software engineering work. Anthropic’s documentation describes a Slack integration that can route coding-related requests from Slack to Claude Code, for teams that have the necessary access, so that a bug report in a thread can lead to a coding session. It requires the Claude app in Slack and an owner enabling Claude Code on the web. Coverage describes Claude Tag as the evolution of this idea, made more proactive and team-oriented, and notes that Claude Code in Slack runs under the requester’s own account while Claude Tag runs under an admin-provisioned identity.
The Slack connector
A connector is a link that lets Claude use another app’s information, with your permission. The Slack connector lets Claude, in its own apps, search and reference relevant Slack messages during a conversation. This is the reverse of Claude Tag: instead of Claude living inside Slack, you bring Slack’s information into a Claude conversation. Anthropic’s page describes this connector as available for Team and Enterprise plan customers who have installed Claude in the Slack app.
| Feature | Direction | Best for |
|---|---|---|
| Claude Tag | Claude works inside Slack channels | Delegating team tasks in the open |
| Claude Code in Slack | Slack requests routed to Claude Code | Coding tasks started from a thread |
| Slack connector | Claude apps read Slack content on request | Researching Slack discussions from a Claude chat |
One option puts the helper in your meeting room. Another gives a specialist engineer a door into the room. The third lets your helper in another building read the room’s noticeboard when you ask. Each connects the same two places in a different direction.
Anthropic’s older help article warns that integrations are tied to your Claude account, not to a particular Slack workspace, which can cause unexpected results if you use several workspaces. Read the current documentation if this applies to you.
11Trust, Privacy and Safe Use
A teammate who can read channels and use tools needs sensible ground rules.
Anthropic’s materials note that Claude in Slack keeps the same security standards and privacy settings as your Claude account, with administrators controlling app approval. In the original beta announcement, Anthropic said that Claude only saw messages in threads where it was explicitly mentioned or in direct messages, and that it did not use that data to train models. For the newer Claude Tag, check the current documentation for what Claude can read in channels and how data is handled, since the design is different.
There is one safety point worth understanding clearly. Anthropic’s Claude Code in Slack documentation warns that Claude may follow directions from other messages in the context, so people should only use Claude in trusted Slack conversations. This is a form of what is called prompt injection, where text inside a message, document or web page tries to instruct the AI. In a busy channel, a message from an outsider, or a pasted document, might contain such text. Keeping Claude to trusted channels and reviewing its work reduces the risk.
Use trusted channels
Enable Claude where members and content are trusted, not in open channels with outsiders.
Least access
Give Claude only the tools and data it needs for the channel’s purpose.
Review before acting
Check Claude’s output before it goes to customers, production systems or leadership.
Mind sensitive material
Avoid posting passwords or confidential details in channels where Claude is active.
A diligent assistant who follows written notes might be fooled by a forged note slipped under the door. That is why you keep the door locked to strangers and double-check unusual instructions before acting.
If you use Slack through your employer, follow your organisation’s policies on AI tools and confidential data, and ask your administrators how Claude is configured.
12Working Well Together: Team Habits
A shared teammate works best when everyone treats it sensibly.
Use DMs for drafts and private work
Anthropic’s guidance suggested keeping Claude DMs for draft work before sharing in channels.
Keep requests in threads
Using one thread per task keeps progress easy to follow and context tidy.
Name an owner
Make clear which person is responsible for reviewing and accepting Claude’s output.
Correct in the open
When Claude gets something wrong, say so in the thread, so it and your teammates learn from it.
Agree what Claude may do alone
Decide as a team which tasks Claude can complete on its own and which need approval first.
Review usage and costs
Where controls exist, such as limits and logs, have an owner check them regularly.
Every well-run team agrees how a new colleague fits in: who they report to, what they can decide and when they should ask. A few agreed habits make an AI teammate just as easy to work with.
If Claude returns an empty or failed response, Anthropic’s help pages suggest simply sending the message again, and contacting support if the problem persists.
13Common Mistakes
These errors appear often, and each has an easy fix.
Problem
Following an old guide for the earlier Claude app and wondering why the steps do not match.
Why It’s Harmful
The earlier app was replaced by Claude Tag, so panel-based steps and per-user setup may no longer apply.
Correct Approach
Check the date of any guide and use the current Claude Tag documentation.
Problem
Giving Claude broad access to everything on day one.
Why It’s Harmful
Excess access increases the damage if Claude misreads a request or is misled by text in a message.
Correct Approach
Start with a few trusted channels and limited tools, then widen access as trust grows.
Problem
Tagging Claude with a vague one-line request in a busy thread.
Why It’s Harmful
Claude may guess the goal, scope or audience wrongly, and teammates may be confused about what was asked.
Correct Approach
State the goal, point to the context, describe the result and name any limits.
Problem
Accepting Claude’s output without review because it came quickly.
Why It’s Harmful
Mistakes can spread to customers, systems or leadership, and visible work in a channel can create false confidence.
Correct Approach
Name a reviewer, check facts and figures, and require approval for anything high-stakes.
Problem
Using Claude in channels that include outsiders or untrusted content.
Why It’s Harmful
Text in messages or documents may try to steer Claude, and sensitive information could be exposed to people who should not see it.
Correct Approach
Restrict Claude to trusted channels and keep sensitive work in controlled spaces.
Problem
Assuming Claude can see everything you can see.
Why It’s Harmful
Claude uses the access it was given, which may be narrower than yours, so answers may be incomplete without warning.
Correct Approach
Link or paste the key context, and ask admins about the scope of Claude’s access.
14Benefits and Trade-offs
Here is the whole picture in one place.
Advantages
- Help arrives inside the tool where the team already talks
- Work is visible, so teammates can follow and correct it
- Long or multi-step tasks can run while people do other things
- Shared context can build up in a channel
- Admins can scope access and apply oversight controls
Disadvantages / Trade-offs
- Needs careful setup and ongoing access management
- Shared visibility is not suited to private matters
- Output still needs human review
- Risk from misleading text in channels and documents
- Product details and plans are still changing
A capable new hire can transform a team, provided they are onboarded properly, trusted step by step and have their work checked at first. The same careful approach brings the best out of Claude in Slack.
15Frequently Asked Questions
It is Anthropic’s newer way for teams to work with Claude, beginning with Slack. Claude acts as a shared teammate that anyone in allowed channels can tag with @Claude, working in the open and reporting back in the thread.
Anthropic’s help pages say that Claude in Slack was switched over to the new Claude Tag experience on August 3, 2026, and that people integrating Claude and Slack should use Claude Tag.
An administrator or owner. Slack admins approve the Claude app, and coverage describes a Claude owner configuring Claude Tag in the admin settings, choosing channels, tools and data. Check the official documentation for exact steps.
In shared channels, coverage says Claude uses an identity provisioned by administrators, not your personal account, with access scoped per channel. More sensitive personal tools are described as staying in DMs under your own identity.
The original beta said Claude only saw messages in threads where it was mentioned or in DMs. For Claude Tag, access depends on admin configuration, so check the current documentation and ask your administrators.
No. Claude Code in Slack routes coding requests to Claude Code under your own account. Anthropic describes Claude Tag as an evolution of Claude Code made more proactive and team-oriented, running under an admin-provisioned identity.
It lets Claude in its own apps search and reference relevant Slack messages during a conversation. It is described as available for Team and Enterprise plan customers who have installed Claude in Slack.
Use it in trusted channels. Anthropic’s documentation warns that Claude may follow directions found in other messages in the context, so avoid channels with untrusted content, limit its access and review its work.
Anthropic has said it plans to expand Claude Tag beyond Slack to other collaboration platforms. For now, Slack is where it is available.
Anthropic’s Claude Tag documentation on claude.com and the help articles at support.claude.com are the best sources, along with your Slack workspace’s admin settings. Details have been changing, so verify before relying on them.
16Summary and Key Takeaways
Claude in Slack has moved from a personal helper to a shared teammate. The earlier Claude app let individuals chat with Claude through DMs, a side panel and mentions, drafting privately before sharing. Claude Tag, which replaced it from August 3, 2026, gives a team one Claude identity set up by an admin, so anyone in allowed channels can tag @Claude into work, watch it plan and carry out tasks in the open and receive results in the thread, with reported features such as persistent channel memory and proactive follow-ups. A separate Slack connector lets Claude in its own apps search your Slack content, and Claude Code in Slack routes coding tasks to Claude Code. Because a shared agent can read and act, access matters: use admin-provisioned identities, give the least access that works, keep Claude to trusted channels, review its output and guard against misleading text in messages and documents. Write clear requests, agree team habits, and remember that this product is evolving, so check the current documentation.
Key Takeaways
- Three connections — the earlier Claude app, the newer Claude Tag and the Slack connector each link Claude and Slack in a different way.
- Claude Tag is the current product — it replaced the earlier app from August 3, 2026, and acts as a shared teammate tagged with @Claude in allowed channels.
- Multiplayer is the point — work happens in the open in a thread, so teammates can follow, correct and build on it.
- Admins control access — a Claude identity and per-channel scope decide what tools and data Claude can use, separate from your personal access.
- Write clear requests — state the goal, link the context, describe the result and set limits.
- Use it safely — stick to trusted channels, grant least access, review outputs and beware of instructions hidden in messages or documents.
- Check current details — plans, settings and features are still changing, so rely on Anthropic’s documentation and your admins.