Claude Tag: What It Is, Pricing, and Safe Slack Setup

Claude Tag puts one shared Claude inside each enabled Slack channel, but the buying decision turns on 3 surfaces, 2 billing identities, and one default most teams should change before launch. Start with one channel, make it mention-only, cap its usage, and grant read-only access to the smallest useful tool set.

Claude Tag is not another private chatbot tab. It is Anthropic's beta Slack agent for Team and Enterprise organizations, designed to keep channel context, take tasks from anyone in the channel, and use company tools under a controlled service identity.

The upside is shared operational context. The risk is that shared context, proactive replies, tool access, and organization-funded usage all meet in the same Slack channel. That makes setup policy part of the product, not a task to postpone until after the pilot.

Claude Tag: What It Is and Why Shared Context Matters

Claude Tag gives a Slack channel one persistent Claude rather than creating a separate assistant for every employee. Everyone in that channel can see its thread, steer the work, correct it, or continue where a colleague stopped. Anthropic says its memory is scoped by channel and workspace, so a sales channel and an engineering channel can hold different context and permissions.

That multiplayer model is the reason to consider it. A support lead can ask Claude to summarize a long incident thread, an engineer can add a missing constraint, and the lead can request the final ticket without re-explaining the conversation. The work stays legible to the people already responsible for it.

It is also the first governance boundary to understand. The channel's Claude is shared, so a correction, saved instruction, connected repository, or overly broad permission can affect work requested by other members. Treat each enabled channel as its own operating environment.

Claude in Slack Has Three Surfaces

The surface changes who sees the work, whose tools Claude can use, and who pays.

The direct-message rule has one important exception. Anthropic says a DM from someone without a connected Claude account can be charged to the organization if that option is enabled. Owners should review that setting instead of assuming every private conversation stays outside the company budget. Anthropic's current help page explains the identity split, and its spend documentation defines the billing paths.

Claude Tag for Teams and Enterprise

Claude Tag is in public beta for Claude Team and Enterprise customers on Anthropic's first-party service. It is not available on Free, Pro, or Max plans, and it is not a feature of third-party Claude deployments. Slack is available now; Microsoft Teams is still marked coming soon on the product page.

Eligibility does not mean every organization can switch it on. Anthropic's setup prerequisites exclude organizations using Zero Data Retention or customer-managed encryption, require routines to be enabled, and require funded usage on Team. Only a Primary Owner or Owner can configure Claude Tag's access and channels. The Claude Admin role cannot do the setup.

What Changed in Late September 2026

The launch is old enough to have a stable definition and new enough that operating behavior is still moving. Anthropic's release notes record the original Team and Enterprise Slack launch on June 23, 2026. Two later details should change how an operator pilots it.

First, personal connectors began rolling out to Team plans on September 24, 2026, with Enterprise to follow. A person can ask the shared channel Claude to use their own calendar, drive, CRM view, or other connected source for that request. Other channel members cannot invoke that person's connector, and the requester can review the response before it posts. The result itself is still visible to the channel, so connector privacy does not make the posted answer private.

There is another boundary: personal connectors do not run unattended. Scheduled routines and work Claude starts on its own use the shared connectors an Owner attached to the channel. That keeps background work under the service identity instead of silently borrowing an employee's access. Anthropic describes those boundaries in its September 24 personal-connectors update.

Second, the proactive-reply control now deserves a setup step of its own. Anthropic's June launch post described ambient behavior as available "if enabled." When verified against the live documentation on September 30, 2026, Respond automatically was on by default for each channel, with no workspace-wide or organization-wide switch. The release notes do not assign a date to that default, so it is safer to treat this as the current operating state, not as a dated launch claim.

Claude Tag Pricing: Usage, Not Another Seat

There is no separate Claude Tag per-seat price to quote. Access requires a Team or Enterprise plan, while channel work is consumption-based and draws from the organization's usage balance. Anthropic also says there is no published per-task cost guidance, so any universal estimate per ticket, thread, or employee would be guesswork.

The practical budget model is a hard organization-wide monthly cap plus a smaller cap for each enabled channel. Alerts fire at 75% and 95% of either limit, and work that would spend beyond a hard cap is refused rather than allowed to run past it. The usage page breaks spend down by channel.

Consider a deliberately bounded example, not a default recommendation. With a $500 organization cap and a $100 pilot-channel cap, that channel can consume at most 20% of the total budget. Channel alerts arrive at $75 and $95; organization alerts arrive at $375 and $475. Those thresholds give the Owner time to inspect which work is consuming budget before the hard stop.

Not every message costs money. Anthropic says reading a channel, deciding whether to answer, and a short reply from context Claude already knows are not billed. A working session is billed, including a proactive session Claude starts without being tagged. That distinction matters when ambient mode is active: passive awareness is not the spend risk, autonomous work is.

For broader context on Claude's other plan and usage limits, see Claude Code pricing in 2026. The billing model here is different enough that its numbers should not be copied into a Claude Tag budget.

What It Means for Operators, Builders, and Buyers

Operators get a shared queue, not a private assistant. Claude can catch up on decisions, turn discussion into work, and preserve context between requests. That is useful in an incident, support-escalation, or launch channel where several people need to inspect and steer the same output. It is a poor fit for a general social channel because saved context and proactive replies will be less predictable.

Builders and security leads own the blast radius. A channel session runs with the organization's service credentials in an Anthropic-hosted sandbox. Give Claude a separate account in each tool, prefer read-only roles, grant specific repositories rather than an entire organization, and keep destructive actions out of the first pilot. The service identity makes access revocable and actions auditable without borrowing an employee's password.

Buyers need to budget the work, not the headcount. Channel users can work with Claude without every participant holding a Claude seat, but channel sessions spend the shared usage balance. A cap tells finance the maximum exposure; per-channel analytics reveal whether the useful work comes from engaged, proactive, or scheduled sessions. This is more controllable than an uncapped agent, but less predictable than a flat software seat.

If the wider decision is whether Claude fits the company at all, the current Claude review covers the core product. Claude Tag should be evaluated as a governed shared-agent layer on top of that product, not as a replacement for every Claude workflow.

How to Use Claude in Slack Safely

The safest setup is a constrained operating trial with explicit ownership. It should prove one useful workflow before it earns more channels or write permissions.

1. 1

   Confirm the organization is eligible

   Use a Team or Enterprise organization on Anthropic's first-party service. Confirm routines are enabled, usage is funded where required, and the organization is not using an incompatible ZDR or CMEK policy.

2. 2

   Put the right owners in the room

   A Claude Primary Owner or Owner must run Claude Tag setup, and a Slack workspace admin must approve or pair the app. A Claude Admin alone cannot complete it.

3. 3

   Pair Slack and scope the pilot

   Install the Claude app, pair the workspace from Claude Tag admin settings, and enable one private or low-risk channel. Do not begin with the entire workspace just because setup offers that scope.

4. 4

   Create a narrow service identity

   Give Claude separate accounts for the tools it needs. Start with a documentation source and issue tracker in read-only mode. Leave email, production systems, broad repositories, and customer records disconnected until the workflow proves it needs them.

5. 5

   Set both caps and alerts

   Set an organization-wide monthly cap, then a lower limit on the pilot channel. Name the Owner who will respond to the 75% and 95% alerts and review the per-channel usage report.

6. 6

   Make the channel mention-only

   Turn Respond automatically off for the pilot channel. Test a summary request, a source lookup, and a draft ticket, then confirm the tool audit logs show Claude's service identity.

Limit Ambient Mode Before It Limits Attention

Ambient mode means Claude can read channel context and speak up without an @mention. The current response-control documentation says Respond automatically is on by default, but separately for every channel. There is no master organization switch.

In Slack, say @Claude only respond in this channel when someone @-mentions you directly. The same control appears on the channel's Configure page and in Claude Tag admin settings. If Claude is already in a thread, mute that thread separately. Later, turn automatic responses on only where you can write a narrow instruction such as answering deploy-process questions or triaging a specific queue.

The rule is simple: use mentions for general collaboration, and reserve proactive behavior for a channel with a named job, named owner, and measured budget.

Claude Tag vs Claude Code in Slack

Claude Code runs as your coding agent with your credentials, local or web session, and developer configuration.
Claude Tag starts from Slack in an organization-controlled sandbox, using channel credentials, shared memory, permissions, and budgets; Anthropic maps the differences in its Claude Code user guide.

Who Should Turn It On Now, Who Should Wait

Turn it on now if one channel already contains recurring, visible work that several people coordinate, such as incident triage, support escalation, launch tracking, or an internal documentation queue. The best first task has clear source material, an output a human can review, and little downside if Claude declines or needs correction.

Wait if your organization needs ZDR or CMEK, cannot fund usage with a hard cap, lacks an accountable Owner, or would have to grant write access to sensitive systems on day one. Also wait if the desired workflow lives mainly in DMs. A shared channel agent adds the most value when its context and output should be visible to a group.

Most knowledge workers are unaffected until an Owner enables their workspace and channels. Microsoft Teams users are also unaffected for now because Anthropic still lists that integration as coming soon.

What's Overhyped

Claude Tag is not a flat-rate digital coworker. Long investigations, tool use, scheduled work, and proactive sessions consume a shared usage balance, and Anthropic does not publish a standard cost per task.

It is not privately personalized inside a channel. The Claude is shared, its context can be steered by other members, and anything it posts can be read by the channel. Personal connectors preserve who may invoke an account, not secrecy for the resulting post.

Ambient presence does not mean Claude is constantly doing paid work. Reading and monitoring are unbilled, and the usual response to an irrelevant message is nothing. The operational concern is a working session started proactively, which is billed and can distract the channel even when the cost is contained.

Finally, it is still a beta. Controls, defaults, eligibility, and connector behavior can change. The late-September documentation already differs in emphasis from the June launch copy, which is why an Owner should review live settings before every expansion.

Your First Week: One Channel, One Cap, Two Read-Only Tools

Pick one private or low-risk channel with a real backlog. Give its shared Claude read-only access to the team's documentation and issue tracker, and nothing else. Set a per-channel cap inside the organization cap, assign one Owner to watch the 75% and 95% alerts, and keep Respond automatically off.

Use @Claude for three bounded jobs: summarize a thread with source links, find the relevant internal procedure, and draft an issue for human approval. At the end of the week, review answer quality, tool logs, proactive versus engaged usage, and spend. Expand only if the channel saved coordination time without creating access ambiguity or noise.

What is Claude Tag?

Claude Tag is Anthropic's beta shared-agent experience in Slack for Team and Enterprise organizations. People can use @Claude in channels, send it DMs, or open it in Slack's AI assistant panel.

How to enable Claude tag?

A Claude Primary Owner or Owner starts setup in Claude Tag admin settings, pairs the Slack workspace with help from a Slack admin, chooses tools and channels, sets a spend limit, and launches it. A Claude Admin role alone cannot enable it.

How much is Claude Tag?

Claude Tag has no separate published per-seat price. It requires Team or Enterprise and charges channel work against the organization's usage balance, controlled by organization-wide and per-channel caps.

How do I use Claude Tag?

Mention @Claude in an enabled channel to guarantee a response, send a DM for personal-account work, or use the assistant panel on the right side of Slack. Keep the first channel mention-only until its context, access, and spend are understood.

Can I use Claude Tag with Microsoft Teams?

Not yet. Anthropic's product page says Slack is available now and Microsoft Teams is coming soon.

What does 'claude tag anthropic' mean?

It is a search phrase for Claude Tag, Anthropic's hosted Slack agent. It is distinct from an unofficial Slack bot built with the Claude API.

Who can use the anthropic claude slack integration?

Claude Team and Enterprise organizations can use the beta in Slack when a Primary Owner or Owner completes setup. Access for individual members then depends on the organization's member and channel settings.

For one practical operating decision like this each week, join the newsletter.