- Channel, group DM or one-to-one DM: whose access, whose bill
- Who sets it up (and why your Admins can't)
- Credits, spend limits and the two alerts
- What happens inside a working thread
- Standing work, memory and the Activity page
- Rolling out Claude Tag, step by step
- My verdict on Claude Tag
- Frequently asked questions
Claude Tag is how Claude works inside Slack as of October 2026: you tag @Claude into a conversation and it takes on real work using your organisation's tools and the context around it, under its own identity. It replaced Claude in Slack on 3 August 2026 and is in public beta on Team and Enterprise plans.
It remembers relevant information from the channels it's in and can follow up on its own. You can reach it three ways: by tagging it in a channel, in a direct message, or through the AI assistant panel in Slack's header. Anthropic's product page says Microsoft Teams is coming soon, with a waitlist. The part that matters most for an admin is that channel work and personal work are billed and permissioned differently.
Channel, group DM or one-to-one DM: whose access, whose bill
When you tag @Claude in a channel, the whole exchange stays visible to everyone there, and anyone can steer it or pick up where it left off. In a channel, Claude acts under your organisation's identity, with the tools and access an admin set for that channel, and the work is billed to the organisation. In DMs and the assistant panel, it uses whatever you've enabled in your own Claude account.
| Where you tag Claude | Tools and access it uses | Who's billed | Best for |
|---|---|---|---|
| Team channel | The channel's connections, set by an admin | Your organisation | Shared work the team should see |
| Group DM | Workspace and organisation connections | Your organisation | Shared work among a few people |
| One-to-one DM | Your own claude.ai connectors | Your own Claude account, if you've connected one | Personal tasks on your own data |
Anthropic's docs draw the line against its other tools too. Claude Tag runs in Slack on the team's service-account credentials, so everyone in the channel sees the work. Cowork runs in claude.ai chat on your personal connectors, and Claude Code runs in your terminal or IDE on your local credentials.
Who sets it up (and why your Admins can't)
Only a Primary Owner or Owner can set up Claude Tag's access and channels. The Admin role can't. Once the Claude app is installed in Slack, the Owner provisions Claude's identity, connects the organisation's tools and repositories, and chooses the channels it can work in. Team members don't need to set up anything themselves.
- Member Access has three modes: anyone in the Slack workspace, any member of your Claude organisation, or only members whose role allows it. The role-based option is Enterprise only, through a Claude Tag in Slack capability on a custom role.
- Access is set at three levels that inherit from the one above: organisation-wide, workspace (all public channels in a Slack workspace) and private channel. A private channel can hold extra credentials, so a legal channel keeps its tools and memory apart from an engineering one.
- Channels with Slack guests have an Allow Claude to respond to guests setting with three choices: Restrict, which is the default and blocks Claude from those channels, Channel only, or Allow.
Credits, spend limits and the two alerts
Claude Tag is consumption-based, so spend follows usage, not the number of people. On a Team plan the spend limit step is required: a Team plan has no usage balance until it's funded, and Claude won't respond in channels until then. You buy credits in the usage section of claude.ai's admin settings. If you're still weighing up seats, my guide to the Claude Team plan explains what each one includes.
On an invoiced Enterprise plan the limit is only recommended, but I'd treat it as essential, because usage bills to the invoice with no upper bound until you set one. The controls are an organisation-wide hard cap, optional per-channel limits (new channels inherit a default) and threshold alerts. The first alert fires at three quarters of the cap and a second one comes nearer the cap; Anthropic's support article and admin docs give slightly different figures for that second warning. Work that would go over a limit is declined, never silently cut short.
- The limit resets at the start of each month and covers every paired workspace.
- It counts usage at list price. A negotiated discount applies at invoice time, not to the cap.
- Reading a channel or group DM, deciding whether to reply and short replies from what Claude already knows aren't billed. Group DM work bills the organisation.
- There's no published per-task cost guidance. Anthropic suggests setting a first-month limit you're comfortable with, then watching the per-channel breakdown.
- To charge spend back to teams, map each channel to one team and use per-channel limits as team budgets.
- Separate throughput limits can make Claude say it's hit a rate limit and name a short wait of a few seconds. A rate-limited request spends nothing.
What happens inside a working thread
Typing @Claude with a task starts a working session in a thread under your message. An isolated, temporary cloud sandbox is built for that thread, Claude works with the channel's access, and the result lands in the thread. The sandbox is then released, but the thread stays, and a new reply rebuilds it.
Longer tasks get a live checklist that Claude edits in place. Slack doesn't notify on edits, so a quiet thread usually means it's mid-task, not stuck. Anyone in the thread can steer by replying, without mentioning @Claude again, and the Stop button pauses the work while keeping the session. Results come back as a reply, a file or chart in the thread, a page kept current by editing, or a hosted page on claude.ai. For code, it's usually a draft pull request under the Claude GitHub App.
The footer under each reply names the model that handled it. You can ask for a different one in the thread, set a channel default, or use fast mode with !fast on Opus models, at a higher cost per token, once an Owner allows it.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Standing work, memory and the Activity page
You can give @Claude standing instructions: watch a channel, run a weekly digest, flag anything urgent or page the right person. Ask @Claude what triggers do you have set up here? in any channel to see and switch off standing work. Owners can open the Activity page, under Organization settings, then Claude Tag, then Activity, which lists every scheduled and one-time task and every network call made with Agent Identity. Pull requests show the Claude GitHub App as author, with a link back to the Slack thread.
Slack conversations with Claude stay separate from your Claude chat history. If you disconnect the integration or uninstall the app, those conversations are deleted from Claude within 30 days. Channel and workspace memory is kept, and Owners can view, edit and delete it.
Rolling out Claude Tag, step by step
- An Owner or Primary Owner installs the Claude app in Slack and provisions Claude's identity.
- The Owner connects the tools and repositories Claude may use and picks the channels.
- On Team, buy credits and set the monthly spend limit. On Enterprise, set one to cap the pilot.
- Choose Member Access: the whole workspace, your Claude organisation, or role-based on Enterprise.
- In a channel's message box, not inside a thread, type /invite @Claude, then ask: @Claude what can you access from this channel?
- Hand it a real task, watch the checklist and reply in the thread to steer.
My verdict on Claude Tag
I'd start with two or three channels where the work's already shared, each mapped to one team with its own channel limit, so you can see who's spending what. I'd leave the guest setting on Restrict until you've decided what outsiders should see. And on Enterprise, I wouldn't let anyone tag Claude until a hard cap exists, because without one there's no ceiling at all. If you're also reviewing which Slack plan your team needs, I've written a guide to what each Slack plan costs as well.
If you're about to roll Claude Tag out to a team and want it used well from week one, that's the kind of thing I help with. I run AI training and workshops for teams, and help businesses with AI enablement, AI automation and GEO, which is about getting your business seen and cited in AI answers. Most of that work is with SaaS marketing teams, telecom companies and small businesses. Here's how we can work together.
Frequently asked questions
Who can set up Claude Tag?
Only a Primary Owner or Owner. The Admin role can't set up Claude Tag's access and channels, and team members don't need to set up anything themselves.
Who pays when someone tags Claude in a Slack channel?
Channel work and group DM work are billed to the organisation. A one-to-one DM bills the sender's own Claude account if they've connected one.
Why isn't Claude replying in our Slack channels?
On a Team plan, Claude won't respond in channels until the organisation has bought credits and set a spend limit, because the plan has no usage balance until it's funded.
Is Claude Tag the same as Claude in Slack?
Claude in Slack switched over to the new Claude Tag experience on 3 August 2026, so Claude Tag is now the way to use Claude in Slack.
