What Is Claude Tag? How to Hand Work to @Claude in Slack: Usage, Pricing, Setup, and Why Personal Pro and Max Plans Can't Use It

Claude Tag is an Anthropic feature that lets you hand work to Claude just by typing @Claude in a Slack channel. Claude researches inside the thread, works through a checklist it updates as it goes, and posts back answers, files, web pages, or draft GitHub pull requests. It was announced on June 23, 2026, and is still in public beta.

Working from the original text of the official Claude Tag documentation, this article covers how it works, how to tell whether you can use it, where the cost is billed, real examples of requests, scheduled runs, how to connect external tools and GitHub, and how it behaves in channels with guests and with your data. Everything reflects the docs as of October 4, 2026. Because it's a beta, details may change before general availability.

1. What is Claude Tag? A shared team @Claude that works in threads

Claude Tag isn't a bot that each person connects with their own account. Instead, the organization has one @Claude that the whole team shares. The organization's Owner sets up dedicated Claude accounts in each tool, such as the issue tracker and GitHub, so anyone in a channel can hand off work without configuring anything. The work runs in the cloud, so it continues even after you close Slack (How Claude Tag works).

Each request follows what the official docs describe as a repeating cycle of five stages.

The working environment uses the same mechanism as Claude Code on the web. That's why what comes back is something that actually runs, not just chat. The flip side is that files that exist only inside the environment disappear once it goes quiet. For long tasks, the official recommendation is to ask Claude to "push to a branch along the way" or "post drafts as you go."

What you see while it works, and how to step in

• The checklist updates by editing the message. Slack doesn't notify on message edits, so even if a thread looks stalled, opening it often shows items have progressed.
• Anyone in the channel can redirect the work just by replying in the thread. It doesn't have to be the person who asked, and there's no need to mention @Claude again.
• To stop it, press Stop on the in-progress indicator under the thread. The session is kept, so you can resume with @Claude (DMs have no Stop).
• Editing an earlier message doesn't change the work. An edit arrives only as a note with the before-and-after text, and a deleted reply isn't removed from the record. Put corrections in a new reply.

What comes back

Source: "What Claude posts back" in How Claude Tag works (checked October 4, 2026)

These web pages are the same thing as Claude Code artifacts, but who can view them is decided differently. Anyone who can join the originating channel can open them, and there are no sharing settings. Assume a page made in a public channel can be opened by everyone in the workspace (Security and data handling).

Even without connecting a single external tool, Claude can read the thread and channel history (including pins), search the workspace, and write and run code inside its environment. That alone is enough to build a chart from a pasted CSV or turn a long thread into a document.

2. Can I use it? A decision flow, and why personal Pro and Max plans get stuck

Whether you can use it depends not on your own Claude plan but on whether your Slack workspace is connected to a Team or Enterprise organization. Work through the questions from the top.

The "no seat needed" point in Q1 is easy to miss. The official overview says Slack users don't each need a Claude seat to use Claude in channels, because channel work is paid from the organization's balance. Owners can restrict access to "only people with an organization Claude account" on Team, or to "only people in specific roles" on Enterprise.

For individual plans, the official overview states plainly that they're excluded.

"It isn't available on individual plans (Free, Pro, or Max)"
—Claude Tag overview

What happened when I clicked it

On October 4, 2026, this site's operator opened claude.ai/code on a personal Max plan and found a prompt in the bottom-left corner inviting them to try Claude Tag, with a setup link. Clicking it opened claude.ai/admin-settings/claude-tag, where, under a heading equivalent to "Set up Claude Tag," the page said to set it up from a Team or Enterprise workspace and that you might be signed in with a personal account. The only two buttons were the equivalents of "View setup guide" and "Go to chat." There was no way to proceed with setup.

This is by design, not a bug. The first row of the table in the official troubleshooting page describes exactly this screen (a page titled "Set up Claude Tag" with "View setup guide" and "Go to chat" buttons) and explains that the signed-in account can't run setup. The fix splits three ways.

Source: the Setup errors table in Troubleshoot Claude Tag setup (checked October 4, 2026)

Nothing in the official docs explains why the prompt appears on individual plans at all. Rather than guess at a reason, I'll simply record the fact: "the prompt appears, but individual plans aren't eligible."

If you want to use Claude from Slack as an individual

Personal Pro and Max plans still have the long-standing Claude Code in Slack. The Claude Code official documentation says that because Claude Tag isn't available on individual plans, that page remains the setup guide for them. The two are quite different, though.

• It's focused on coding requests, launching Claude Code cloud sessions from Slack. It requires a GitHub connection and a one-time setup on claude.ai/code.
• It works only in channels (public and private), not in DMs.
• Usage comes out of the plan allowance of the person who made the request.
• A Slack workspace admin needs to install the Claude app from the Slack App Marketplace.

In Team and Enterprise organizations, this older version is scheduled to be replaced by Claude Tag. For individuals, it is currently the only official route (see section 13 for a comparison).

3. Pricing: which work is billed where

Adding Claude Tag doesn't increase the per-seat price. Instead, the type of work determines which budget pays for it. Here is the table from the official spend limit page, laid out as a flow.

Source: Set a spend limit, Restrict where Claude Tag operates (checked October 4, 2026)

Choosing a monthly limit

The Owner picks the monthly limit on the last setup screen (organizations that went through the credit purchase screen set it on the usage page after launch). The options are:

Source: "Launch Claude Tag" in Set up Claude Tag (US dollars; checked October 4, 2026)

• On a Team plan, nothing runs until you buy usage credits. With a zero balance, asking in a channel won't start any work.
• Invoiced Enterprise organizations have no cap until they set a limit. The official docs recommend setting one even during a pilot.
• There's no published guidance on how much a single task costs. The official recommendation is to set a limit, try it, and check the per-channel breakdown on the usage page.
• The limit counts usage at list price. Even with a negotiated discount, the discount applies at invoice time, not to the cap. The limit resets at the start of each month.
• When the limit is reached, Claude stops working and says so in the thread. Separately, there's a rate limit for too many requests in a short time; waiting a few seconds and re-sending gets through. Raising the spend limit doesn't clear a rate limit.
• You can't set per-person limits. Limits exist only at the organization and channel level.

One-to-one DMs from members who haven't connected a Claude account have the following limits per person. Once any one of them is reached, Claude stops answering at the organization's expense.

Source: "Limits on direct messages from members without a Claude account" in Restrict where Claude Tag operates

Organizations that want to turn off these limited-time DMs can either restrict users to organization members or turn off DMs entirely. Both are Owner settings and affect other uses too (the former also blocks unconnected members in channels, and the latter stops DMs for connected members as well).

The analytics page in the admin console splits channel spend into four categories: work Claude did when called (Engaged), work it started on its own (Proactive), scheduled runs (Scheduled), and just reading the channel (Monitoring, not billed). For channels where you don't want Claude starting work uninvited, turn off the "auto-reply" setting described in section 4.

4. How to ask: channels, DMs, group DMs, and channel memory

Where you ask changes whose tool permissions are used and who can see it. Based on the examples in the official Get started page, requests look like this.

Getting started in a channel takes three steps.

1. Run /invite @Claude in the channel's message box (you can't send it from inside a thread).
2. Ask @Claude what can you access from this channel? If you get a reply, it's working, and the answer also tells you which tools are available in that channel.
3. Send @Claude with a task. Once Claude is in a thread, it responds to later replies without being mentioned.

What Claude can and can't read

Source: "What Claude can see" and "Files you attach" in Get started (checked October 4, 2026)

Check answers before you use them. The official docs also say to match how closely you read to how much is at stake: skim a summary, but read anything going to customers or changing systems carefully. If you're curious about the basis for an answer, you can ask in the same thread, "Which channels did you search?" or "Show me the query you ran."

Channel memory: teaching Claude the rules

If you ask something like @Claude remember for this channel: put reports in a table and always include source links, it's saved to channel memory and applies to everyone's threads in that channel. Memory belongs to the place, not the person (What Claude Tag remembers).

• Check what it remembers with @Claude what do you remember about this channel? Anyone in the channel can have it correct or forget wrong or outdated entries.
• Keep memories short. Capacity is limited, and long entries push others out. Put long runbooks in a repository and link to them.
• Workspace-wide memory learned in a public channel is read in every channel, including private ones. Splitting access permissions by channel doesn't split knowledge.

Tuning how much it replies

Claude reads messages in the channels it belongs to even when it isn't called. However, it usually does nothing with messages that don't address it. It replies only in limited cases, such as offering help once when someone clearly describes a problem, or when it can answer in a single message from history or memory (Control when Claude Tag responds).

• Make the whole channel "only when mentioned": ask @Claude in this channel, only reply when you're @mentioned (this turns off the "auto-reply" setting, which is on by default).
• Silence a single thread: send @Claude !mute in that thread. To undo it, send @Claude !unmute.
• Remove it from the channel: /remove @Claude. Memory and scheduled runs are kept, not deleted.

5. Routines (scheduled runs) and channel monitoring

A request that worked well can become a Routine (scheduled run) that Claude repeats. Setting one up takes a single message in the channel, and it runs with that channel's tools and permissions (Set up routines). Based on the official examples, there are three types.

The official docs recommend spelling out both the channels and the topics to watch for monitoring, and the output format for recurring posts, so they stay readable over time. Pull request tracking is the only way to react to GitHub events; a setup like "every time a new PR is opened" can't be configured from Slack.

⚠️ Schedules run in UTC. Name the time zone when you ask, as in "9am Pacific." If you don't, the time zone on your Slack profile is used. You can check the time that was set with @Claude !routines (the list shows times in UTC). If you schedule in US time or another zone with daylight saving time, the local time shifts by an hour at each clock change.

• It posts only to that channel or to another public channel Claude is in. Posting to a different channel is possible only when the original channel is public. It doesn't send DMs to individuals.
• Routines created in a group DM post only to that group DM.
• A routine doesn't stop when its creator leaves the channel. Anyone still in the channel can list it, change its time, or stop it.
• A recurring job that can't reach GitHub stops itself after failing three times in a row over a span of an hour or more.

6. Choosing a model

Like any other request, you switch models in plain language. The name of the model that answered appears in the footer under each reply, so check there to confirm a switch (Choose the model Claude Tag uses).

Source: Choose the model Claude Tag uses, Claude Tag for Claude Code users (checked October 4, 2026)

You can choose from the Opus and Sonnet families, limited to what your organization allows. If you ask by family name, such as "the latest Opus," instead of a version number, you get the newest model in that family, and if the channel default is set to a family name, it moves to new models automatically when they arrive. The official docs give no criteria for which model to use for which job; the only guidance is to use fast mode where speed matters more than cost (such as an incident-response thread).

7. Setup steps (for admins)

An Owner does the setup once, and everyone in the channels can use it from the moment it launches. First, confirm you have these seven things (Set up Claude Tag).

Source: the prerequisites table in Set up Claude Tag (checked October 4, 2026)

The Owner starts by clicking Start setup at claude.ai/admin-settings/claude-tag. If you leave partway through, Resume setup picks up where you left off.

If your Slack admin is someone else, the official page includes a request message you can send to the admin as is (asking them to install the app and send back the code). The app permissions admins care about are summarized in What the Claude Slack app can access. Installing the app alone doesn't put Claude in any channel.

After launch, you make changes on the same admin page under Claude Tag's access → the Slack tab. There are three levels, Default Slack (the whole organization), workspace, and channel; what you add at a higher level applies below it, and what you add at a lower level applies only there, on top.

8. Connecting external tools: Access bundles and dedicated Claude accounts

Setup connects only Slack. To let Claude use your issue tracker or databases, you connect tools after launch. There are two routes (Give Claude access to your tools).

• Access given to Claude: an admin connects a tool with credentials that belong to Claude, not to a person. They're used for everyone's requests in that channel and for scheduled runs.
• Members' personal connectors: connectors you've added to your own claude.ai (calendar, email, and so on), used only for your own requests. No admin work is needed.

Connecting with a dedicated Claude account

An Access bundle packages credentials, allowed domains, repositories, plugins, and instructions together. For example, if you split them into three bundles, "read data only," "can write to GitHub," and "monitoring," you can mix and match: all three for an engineering channel, just one for an analytics channel. The official docs recommend connecting tools in these six categories.

Source: "Decide what to connect" in Give Claude access to your tools (checked October 4, 2026)

Source: Security and data handling (checked October 4, 2026)

That's why the rule is never reuse keys from a person's own account. With a dedicated account, the tool's audit log records actions under Claude's name, and you can revoke access without affecting anyone else.

Letting Claude use your own connectors in a channel

When a task you asked for in a channel needs something like your calendar, Claude shows a prompt only you can see and lets you choose from three options (Personal connectors in channels). It's available from the start on Team plans; on Enterprise, the admin page shows whether it's available.

Other people's requests and scheduled runs never use your connectors. However, the official docs describe the sensitivity check as a filter, not a guarantee. For sensitive topics such as HR or customer information, choosing Allow with review is the safer option.

9. The pull request flow on GitHub and GitLab, and two-person approval

To hand off coding work, connect the Claude GitHub App. Only someone who is both an owner of the GitHub organization and an Owner of the Claude organization can set it up. Repositories are allowed per Access bundle (Configure GitHub access). The flow from request to merge looks like this.

⚠️ The person who asked for a Claude PR can approve and merge it alone. GitHub's rule that you can't approve your own PR applies to Claude as the author, not to the person who asked. The official docs recommend requiring two approvals on branches where Claude opens PRs, or creating a check that requires two approvals only for Claude's PRs. Also enable the setting that dismisses stale approvals when new commits are pushed.

GitLab works differently: instead of an app, you put an access token for a dedicated Claude GitLab user into an Access bundle. Claude can manage issues, comment on merge requests, and check pipelines, but because it's an API-only connection, it can't clone or push private projects (you get a 401). The official docs say to connect through GitHub if you want code cloned into the environment (Configure GitLab access). Self-hosted GitHub Enterprise Server and GitLab are supported only if reachable from the internet.

10. Behavior with guests, Slack Connect, and shared channels

Claude is designed to be cautious where people from outside the company are present. Here are the default behaviors and whether you can change them.

Source: "Restrict guest channels," "Slack Connect channels," and "Group DMs" in Restrict where Claude Tag operates (checked October 4, 2026)

Under any setting, guests can read what Claude posts in the channel. With Channel only, treat instructions written in the channel as things guests might also see through Claude's replies. If a channel you were already using becomes a Slack Connect channel, Claude stops answering even in threads it had joined.

11. Data retention and deletion

⚠️ During the beta, Claude Tag data has no automatic retention period. Session transcripts, memory, scheduled runs, and so on remain until you delete them. Retention settings your organization has set don't apply to transcripts and memory either (they do apply to published web pages). A plan lapsing or Claude Tag being turned off doesn't, by itself, delete anything.

What Anthropic keeps is a "record of Claude's work," separate from the Slack messages themselves. It consists of session transcripts (the messages Claude was shown and their senders, attachments, pre-edit text, data retrieved from searches and connections, and Claude's own statements and actions), memory, scheduled runs, each place's settings, Access bundles, the links between Slack and Claude accounts, and published web pages (Data lifecycle and deletion).

Source: excerpted from the per-action table in Data lifecycle and deletion (checked October 4, 2026)

• There's no way to delete the record of just one thread. During the beta, it isn't included in organization data exports, and the Compliance API can't list or delete it. Send other deletion requests to your account team or privacy@anthropic.com.
• Even after deletion, copies remain in standard backups for a certain period.
• It can't be used with protected health information at healthcare organizations. Claude Tag isn't covered by Anthropic's BAA (Business Associate Agreement), and the official docs have a separate page on configuring things so health information doesn't land anywhere Claude can read (healthcare).

12. Common snags and commands worth knowing

These are the issues you're most likely to hit, chosen from the two official troubleshooting pages (for admins and for users).

Sending certain keywords after @Claude triggers a fixed action instead of a normal conversation (Commands Claude Tag understands).

13. How it differs from the earlier Claude in Slack, Cowork, and Claude Code

The Claude app for Slack has an earlier generation (Legacy), and Claude Code in Slack is part of it. Since they're two versions of the same Slack app, there's no need to uninstall anything; you switch between old and new per place in the admin console (migration guide).

Source: Migrate from the earlier Claude in Slack, Claude Code in Slack (checked October 4, 2026)

The retirement date for the old version on Team and Enterprise isn't public; the docs say to "check with your account team." After that date, places still on the old version stop responding. Switching to the new version doesn't carry over the GitHub and other connections each person had set up, so an admin needs to set up Access bundles again.

Choose between Claude Tag and Anthropic's other products by whose permissions it runs with and who can see it. Here is the official comparison table (Cowork itself is covered in What is Claude Cowork, and the differences among the three in Chat vs. Cowork vs. Code).

Source: "How Claude Tag differs from Cowork and Claude Code" in How Claude Tag works

Summary

• Claude Tag lets you hand off work by typing @Claude in Slack. The organization shares one @Claude, which runs in a separate environment per thread. It was announced on June 23, 2026, and is still in public beta.
• Only Team and Enterprise organizations can use it. Getting stuck after clicking the prompt on a personal Pro or Max plan is by design; individuals use Claude Code in Slack instead (channels only, billed to their own allowance).
• Once an organization sets it up, Slack users can ask in channels even without a Claude seat. There's no extra seat fee; channel work is billed by usage from the organization's balance, and on Team nothing runs until credits are purchased.
• Channel memory lets Claude learn the rules, and Routines handle recurring reports, channel monitoring, and PR tracking. Name the time zone in the request.
• Connect external tools through dedicated Claude accounts with the least privilege. Require two approvals on Claude's PRs, and be careful with channels that have guests and with the lack of a data retention period.

FAQ

Q1. If the prompt shows up on my personal Max plan, can't I use it?

No. The official overview explicitly excludes individual plans (Free, Pro, Max), and not being able to proceed to setup after clicking the prompt is the by-design behavior listed in the official troubleshooting page. The official docs don't explain why the prompt appears.

Q2. Can coworkers without a Claude seat use it?

With the default settings, yes. Channel and group DM work is paid from the organization's balance, so Slack users don't each need a Claude seat. If an Owner has restricted access to organization members (or to specific roles on Enterprise), people will be prompted to connect a Claude account.

Q3. How much does it cost per month?

There's no fixed monthly fee. It depends on how much work happens in channels, and no per-task estimate is published. The official recommendation is to set a monthly limit, start small, and estimate from the per-channel breakdown in the admin console.

Q4. Does it work in Microsoft Teams?

As of October 4, 2026, the official documentation covers only Slack. The product page just says "Coming soon to Microsoft Teams," with no release date.

Q5. Are my own Claude Code settings (CLAUDE.md and ~/.claude) used?

Your local ~/.claude, personal settings.json, and locally configured MCP servers are not loaded. Claude Tag runs in a cloud environment the organization provides and can't read your local machine. On the other hand, CLAUDE.md, .claude/rules/, and .claude/skills/ committed to a repository are loaded when Claude clones that repository. However, hooks in the repository's .claude/settings.json don't run, and .mcp.json isn't loaded either. Instead of prompting for permission on the spot, it runs in auto mode, and admins define the permission rules (Claude Tag for Claude Code users).