Anthropic's Claude Tag for Slack Explained

Claude Tag puts Anthropic's Claude agent directly inside a team's shared Slack channels so anyone in a channel can tag it into a problem with no setup of their own. You tag @Claude in a channel or thread, and Claude completes the requested task inside an isolated cloud sandbox using the tools connected to your workspace.

At Layer3Labs, we help businesses evaluate and deploy AI workflows, and with a shared Slack agent the request, the checklist and the result all sit in one thread the whole channel can read.

As of October 2026, Claude Tag is in public beta and replaces the earlier per-user Claude in Slack application. Anthropic's Claude Tag overview says features and behavior may change before general availability.

How Claude Tag Works

You tag @Claude in a Slack message and say what you need. Claude opens a session in that thread, does the work in a cloud sandbox with your team's connected tools, and replies in the same thread.

• 1. Mention: You type @Claude in a channel message or thread and describe the task you want completed.
• 2. Thread session: Claude spins up an ephemeral cloud sandbox running the same execution engine that powers Claude Code on the web.
• 3. Work and checklist: Claude does the task with the tools connected to your workspace. Longer jobs get a checklist it edits in place, and Slack does not notify on edits, so a quiet thread usually means Claude is still working.
• 4. Steering: Anyone in the channel can redirect Claude by replying in the thread, with no second tag. Channel threads have a Stop button; DMs do not.
• 5. Result: Claude replies in the thread with the answer, a file or chart, a page it keeps updated, or a draft pull request.
• 6. Thread edits: Editing a message only sends Claude a note. Deleting a reply leaves it in the transcript, and deleting the thread's first message closes the session.

Core Capabilities and Connected Integrations

Claude Tag operates with native workspace capabilities even before an administrator connects third-party services. Without external integrations, Claude reads channel history, searches the workspace, and writes and runs code inside its ephemeral sandbox environment.

For engineering teams, Claude Tag clones GitHub repositories, creates branches, and pushes changes directly. Code modifications typically return to the thread as draft pull requests opened by the Claude GitHub App. Beyond code, Claude returns results as in-thread messages, downloadable files, visual charts, auto-updating documents, or hosted web pages on claude.ai.

Claude Tag supports Routines, which allow teams to configure scheduled background jobs, automated channel monitoring, and pull request activity subscriptions. The sandbox reaches the network over HTTP and HTTPS only, so a tool you can reach only through another protocol will not connect. Anthropic provides dedicated connector setup guides for Amplitude, Asana, Jira, Confluence, BigQuery, Datadog, GitLab, Gong, Google Drive, Google Calendar, Gmail, HubSpot, Linear, Notion, PagerDuty, Salesforce, Sentry, Snowflake, Stripe, and Vercel, as well as custom Model Context Protocol connectors.

Thread Context and Multi-Channel Memory

Claude Tag reads thread context within a bounded historical window rather than ingesting an entire channel transcript. When tagged into an existing conversation, Claude examines recent messages to establish context while automatically filtering out responses from other Slack bots.

Memory operates across distinct channel and workspace tiers according to Anthropic's how Claude Tag works guide. Each channel keeps its own notes. Notes saved from public channels become workspace notes, which Claude can read across any channel in the organization. Private channels can read these workspace-wide notes but cannot write new notes back to the shared workspace tier.

Channel-level sessions do not persist indefinitely. Anthropic replaces an active channel session after approximately one hour of user inactivity, or automatically once the session reaches roughly one day of total age.

Model Selection and Speed Controls

Claude Tag allows organizations to select which underlying models power their Slack interactions. Users choose among the Opus and Sonnet models their organization allows.

You can pick a model per thread, set a channel default, or set one for DMs. Add the fast-mode command after @Claude for fast mode, which runs on Opus models, per Anthropic's models documentation.

Workspace Plans and Usage Pricing

Claude Tag is available exclusively on Anthropic Team and Enterprise plans on the first-party service, and cannot be added to Free, Pro, or Max tiers. Anthropic does not charge an additional per-seat fee for adding Claude Tag to a Slack workspace.

Work in channels and group DMs bills by usage against the organization's balance. Individual employees participating in channel threads do not need individual paid Claude seats. However, one-to-one direct messages with Claude bill directly to the individual sender's personal seat, which must include a license for Claude Code.

Organizations on the Claude Team plan must purchase usage credits in advance before tasks can execute, though Claude reading channel text without executing a task does not consume credits. Anthropic publishes no fixed per-task cost guidance. Admins set a monthly spend limit during setup (spend-limit options). Claude posts warnings at 75% and 90% of the limit and stops when it reaches it. You can review broader subscription details in our guide on Claude pricing.

Setup Process and Owner Prerequisites

Setting up Claude Tag requires an Anthropic organization Owner working with a Slack workspace administrator according to Anthropic's setup overview. Before starting, verify that your account uses an active Team or Enterprise plan on Anthropic's first-party platform with Routines enabled under Capabilities.

Claude Tag is not available to organizations using Zero Data Retention, customer-managed encryption, or a HIPAA local-mode configuration. Team plans must have active prepaid credits, and the Slack workspace must contain at least one public channel.

After launch, Anthropic recommends setting up a dedicated company email account for Claude. You should invite that dedicated identity into external tools with the narrowest necessary roles, generate API keys under that account, and configure them as connectors.

• 1. App installation: An administrator installs the application from claude.com/claude-for-slack into the target Slack workspace.
• 2. Generate pairing code: Invite the bot by running /invite @Claude in a public channel, then have a Slack admin type @Claude connect to receive a pairing code valid for 15 minutes.
• 3. Pair workspace: Paste the 15-minute pairing code into the Anthropic admin console at claude.ai/admin-settings/claude-tag and click Pair workspace.
• 4. Connect repositories: Link the organization GitHub account if your team intends to use code generation and pull request features.
• 5. Fund balance: Add usage credits to the Anthropic billing balance if your team operates on a Team plan.
• 6. Set spending limits: Select a monthly spending cap of $500, $1,000, $2,500, $5,000, Unlimited, or a custom figure, select approved channels, and click Launch Claude Tag.

Admin Governance and Security Limits

Anthropic provides administrative controls to regulate where and how Claude operates across a company. Admins can deploy Claude Tag across an entire Slack organization, a single workspace, or specific selected channels. Within any enabled channel, every member receives identical access to Claude.

Admins can turn direct messages on or off, designate delegated Claude Tag administrators, set channel-specific spend caps, restrict which channels Claude can search, and restrict access in channels containing guest accounts. The admin console includes an Activity dashboard and channel spend analytics. However, administrators cannot view or export transcripts of individual messages team members send to Claude.

Under the public beta data policy outlined in Anthropic's data lifecycle documentation, Anthropic stores session transcripts, channel memory, routines, configuration settings, credentials, and published artifacts. There is no automated data expiration window during the beta period. Data stays until an Owner disconnects the workspace, the app is uninstalled, the channel is deleted, or its scope is removed. During the beta, Claude Tag data is not in organization exports or the Compliance API. Claude Tag is not covered by Anthropic's Business Associate Agreement (BAA) for healthcare.

Claude Tag Compared with Cowork and Claude Code

Anthropic divides its agent tools between collective team coordination and individual technical work. Claude Tag handles team work in Slack channels, while Claude Cowork and Claude Code handle personal work.

Alternative platforms approach team collaboration differently. In marketing published on July 13, 2026, Kylon argues that Claude Tag uses flat channel permissions because integrations attach to the bot identity rather than individual users. Kylon also contends that Slack lacks structured databases, versioned files, and multi-agent rooms. However, some claims in Kylon's comparison post are outdated compared to Anthropic's current documentation. While Kylon stated that Claude Tag is restricted to Opus models and per-channel memory, Anthropic's documentation confirms that teams can select Sonnet models and share workspace-level notes across channels. Our detailed breakdown of Kylon vs Claude Tag contrasts their deployment architectures.

Claude Tag does not fit teams that need a BAA for healthcare data, organizations using Zero Data Retention, or admins who need to read every message people send Claude. Our assessment would change if Anthropic extends Business Associate Agreement coverage to Claude Tag or integrates session transcripts into the Compliance API. To evaluate Claude Tag today, check the prerequisites in Anthropic's setup overview and run a pilot in a single public pilot channel.