Claude Tag is Anthropic's agent for Slack: you mention @Claude in a channel or thread, it takes on the task with the tools and access set for that channel, and it replies in the thread. This guide covers what Claude Tag does today, how teams use it, where it stops, and when a team needs several named agents instead of one. Runbear, which runs named AI agents in Slack and Microsoft Teams, can deploy an agent you built in Claude Code to Slack for your team: the /runbear:deploy command uploads the project, and /runbear:connect-slack gives the agent its own Slack bot (how to deploy a Claude Code agent to Slack).
What Claude Tag is
Anthropic released Claude Tag as a public beta for Claude Team and Enterprise plans, and it replaces the earlier Claude in Slack app. As of October 2026, it is not offered on individual Free, Pro or Max plans. Each channel has one @Claude that everyone talks to, so the whole channel can follow its progress and anyone can pick up a task a colleague started. You tag it with a request, it splits the work into steps, uses the tools the channel can reach, and reports back in the thread. Anthropic calls it "the beginning of an evolution of Claude Code."
How it differs from the Claude web app
The Claude web app is a separate workspace from where your work happens. You switch tabs, paste context in, and carry the answer back. Claude Tag flips that. It sits in the conversation, sees the thread, and replies in place. It runs the same model, but the channel feeds it context instead of you.
How Claude Tag differs from other Slack bots
Most Slack bots are keyword-triggered and scripted. They match a phrase and return a canned response or a form. Claude Tag reasons over natural language, reads context, and writes a coherent answer that fits the situation. That step up is why teams soon expect it to do more than talk.
How to set up Claude Tag for Slack
An Owner of the Claude organization sets it up, and pairing the Slack workspace also needs a Slack workspace admin. The Owner then launches Claude Tag. Launch includes picking at least one public channel for Claude to join and, in most setups, a monthly spend limit. After launch, the Owner gives Claude its own accounts in the tools it should use (setup guide). Owners decide which tools and data each channel can reach, and they can also cap spend per channel. By default, nobody in Slack needs their own Claude seat to tag @Claude in a channel. The organization pays for that channel work by usage.
What Claude can see, and what it cannot
Claude Tag replies in direct messages and in the channels it has joined. Tools and data an Owner connects for all of Slack apply in every channel, and Claude can search public channels across the workspace, including ones it has not joined. An Owner can narrow both (access controls). In a channel it acts through service accounts an Owner set up, so everyone there gets the same access, whoever asks (agent identity). A member can also allow it to use their own claude.ai connectors for a request they made. Memory works by channel: Claude keeps separate notes for every channel, and in a public channel it can also save workspace-level notes that every channel then reads. Other workspaces and one-to-one DMs stay separate (memory).
How enterprise teams use Claude Tag
Enterprise teams use Claude Tag for conversational, drafting-style work across five functions: support operations, sales operations, engineering, product and design, and leadership.
Support operations
Drafting first-pass customer responses, answering policy and knowledge-base questions, summarizing long support threads, and formatting post-mortems. Updating the ticket or notifying the customer can happen in the same thread when the channel has the support tool connected.
Sales operations
Drafting prospect follow-ups, summarizing discovery-call notes, answering internal questions about pricing and discounting policy, and shaping proposal language from a pasted brief.
Engineering teams
On-demand code review of pasted functions and diffs, first-pass incident triage from error logs and stack traces, documentation drafts pulled out of a discussion thread, and regex or query writing.
Product and design teams
Synthesizing user research feedback, spotting patterns across notes, drafting product requirement documents, and sketching competitive comparisons.
Leadership
Tightening up communications, prepping for board and leadership meetings, and working through decision frameworks.
Claude Tag can carry these from a draft to an action when the channel has the right tools connected. The limit is that each channel has one @Claude, and everyone who asks gets the same access.
What Claude Tag does well, and where it hits limits
Strengths
- Reduces context switching by bringing the model to the conversation.
- Reads thread context, so you skip most of the setup.
- Requires no prompt engineering, since natural language is enough.
- Owners set the tools and spend for each channel, and it replies only in direct messages and the channels it has joined.
Limitations
- One identity per channel. There is one @Claude in a channel. Owners can give different channels different tools, and each channel keeps its own memory, but there are no named agents such as @HR or @SalesOps that people pick by role.
- The same tool access for everyone in a channel. Channel tools run on accounts an Owner set up, so a request reaches whatever those accounts can reach, whoever asks. Per-person access applies when a member lets Claude use their own claude.ai connectors for their own request, and in one-to-one DMs from members who have connected a Claude account, which run on that member's own claude.ai account.
- Slack only, for now. Claude Tag works in Slack. Anthropic says it aims to bring it to more of the places teams work.
- Off in guest channels by default. Claude does not reply in channels that include Slack guests, Slack Connect channels included, until the channel's guest setting is changed (access controls).
- Still in beta as of October 2026. Features and behavior can change before general availability, so check Anthropic's docs before you roll it out.
Why Anthropic uses it internally
Claude Tag is not weak. Anthropic said in June 2026 that its internal version of Claude Tag creates 65% of its product team's code. The product answers to one name, @Claude, in every channel.
Where Runbear comes in
Runbear is an alternative to Claude Tag for teams that want several named agents instead of one @Claude per channel. Runbear runs these agents in Slack and Microsoft Teams, each with its own job, knowledge and tool access. On the Claude Agent SDK runtime, Runbear's usual default, each agent can use Claude, GPT or Gemini.
Teams add Runbear when they want:
- Several named agents. Agents such as @HR or @Platform, each with its own role, knowledge and owner.
- Tool access per person. On plans that include per-user auth, a tool on an agent can be set to Per-User, so it acts through the asking person's own account once they link it. A Shared tool uses one account for everyone. Documents indexed into an agent's knowledge base are available to everyone who can use that agent.
- A knowledge base per agent. Notion, Confluence, Google Drive and other sources synced into the agent.
- Triggers. Schedules, webhooks and events in connected apps, so an agent can start work without being mentioned.
- Edit controls. Owners and Admins can edit every agent, and editing an agent in the dashboard can be limited to the members you invite.
- Usage dashboards. An organization-wide analytics dashboard shows messages, users and feedback across agents.
- Slack and Microsoft Teams. The same agent can serve both. Discord is available on Runbear's other runtimes.
If one shared @Claude per channel covers your team's work in Slack, Claude Tag alone may be enough. A team can also run both. Claude Tag gives a channel one shared @Claude, and Runbear runs named agents behind specific workflows.
What this looks like in practice
Matillion, a data-integration company, runs more than two dozen purpose-built Runbear agents, used by roughly 200 people across the company each month. Its agents carry the company's knowledge, run on triggers, and act in connected tools.
Claude Tag vs. Runbear at a glance
| Capability | Claude Tag | Runbear |
| Models | Anthropic's Claude models | Claude, GPT or Gemini per agent on the Claude Agent SDK runtime, the usual default |
| Where it runs | Slack | Slack and Microsoft Teams on the Claude Agent SDK runtime, Discord on other runtimes |
| Reads thread context | Yes | Yes |
| Shared across the org | Yes | Yes |
| Named agents | No, one @Claude per channel | Yes, several per team |
| Knowledge | Channel and workspace memory, plus the tools connected to the channel | A knowledge base per agent (Notion, Confluence, Google Drive and more), plus connected tools |
| Takes action in external tools | Yes, through tools connected to the channel | Yes |
| Whose access tools use | The channel's service accounts, or a member's own connectors for that member's requests and in DMs from members with a connected Claude account | Per-User tools use the asker's linked account on plans that include per-user auth, and Shared tools use one account |
| Runs without being tagged | Yes: routines, such as scheduled jobs, watching a channel or following a pull request | Yes: schedules, webhooks and events in connected apps |
| Who can change it | Owners, plus Claude Tag admins and channel managers on Enterprise. Channel members can change channel memory, and channel instructions unless an Owner blocks it | Owners and Admins, plus members invited to edit an agent |
| Admin visibility | An Activity page for routines, memory and network events, actions under Claude's service accounts in each tool's audit log, and spend limits | An organization-wide analytics dashboard, and an audit log on plans that include it |
Frequently asked questions
Is Runbear a competitor to Claude or Anthropic?
No. Runbear is an alternative to Claude Tag for running agents in Slack. On the Claude Agent SDK runtime, Runbear's usual default, agents can run on Claude, GPT or Gemini, chosen per agent, and a team can run Runbear alongside Claude Tag.
Do I have to leave Claude Tag to use Runbear?
No. A team can keep Claude Tag for one shared @Claude per channel and add Runbear for named agents, per-person tool access and Microsoft Teams.
Can Runbear take real actions in our tools?
Yes. Through integrations, a Runbear agent can do things like file a ticket, update a CRM record, or send a message, with optional human approval gates.
Does Runbear respect individual access permissions?
Yes, for tools set to Per-User on plans that include per-user auth: the agent acts through the asking person's own linked account in that system. Shared tools use one account for everyone, and documents in an agent's knowledge base are available to everyone who can use the agent.
Does Runbear work outside Slack?
Yes. On the Claude Agent SDK runtime, Runbear's usual default, agents run in Slack and Microsoft Teams, and the same agent can serve both at once. Discord is available on Runbear's other runtimes.
Key takeaways
- Claude Tag brings one shared @Claude into each Slack channel, with the tools, memory and scheduled jobs set up for that channel.
- Its limits today: one identity per channel, the same tool access for everyone in that channel, and Slack only.
- When different teams need their own agents, each with its own knowledge and per-person tool access, Runbear fits better.
- Runbear adds named agents, Per-User or Shared tool access, a knowledge base per agent, triggers, edit controls and usage dashboards, in Slack and Microsoft Teams, with Claude, GPT or Gemini on the Claude Agent SDK runtime.
- To put an agent you built in Claude Code in Slack for your team, Runbear's /runbear:deploy and /runbear:connect-slack commands deploy it and give it its own Slack bot.
Considering Claude for your team? See how it compares with Runbear's shared agents in Slack and Microsoft Teams.
