Claude Tag vs Claude Code: what Claude in Slack remembers

What is Claude Tag?

Anthropic launched Claude Tag on 23 June 2026, in beta for Team and Enterprise customers, and called it “the beginning of an evolution of Claude Code.” Anyone in a channel can tag @Claude with a task. A thread becomes a session in an ephemeral sandbox that runs “the same engine that powers Claude Code on the web” (how Claude Tag works). It clones a repository granted to the channel when a task names it, posts a checklist as it goes, and opens pull requests, usually drafts, under the Claude GitHub App, each linked back to its thread.

It acts under service accounts an admin provisions, not as the person who asked. It replaced the earlier, per-user Claude in Slack on 3 August 2026 (What is Claude Tag?). In a workspace connected to Claude Tag, that legacy bot stopped answering on 5 October 2026. Pro and Max accounts can still use it in a workspace no organization has connected.

Claude Tag vs Claude Code vs Cowork

Anthropic's own comparison, from the how-it-works page:

From a Claude Code setup, only the repository travels. Once Claude clones a granted repository, CLAUDE.md, .claude/CLAUDE.md, .claude/rules/*.md and the skills in .claude/skills/ load (Claude Tag for Claude Code users). Hooks in the repository's .claude/settings.json don't run, and .mcp.json is not loaded even when committed. Nothing on your machine loads: not ~/.claude, your personal settings, your shell environment or your local MCP servers. Sessions run in auto mode, and admins pre-approve routine actions with allow rules.

What does Claude Tag remember?

Memory belongs to the channel, not to you (What Claude Tag remembers). It builds up three ways: you tell it (“@Claude remember for this channel: reports go out as tables”), it saves facts on its own, such as decisions the channel made, and it can read the transcripts of earlier sessions in the channel when you name a timeframe or topic.

In a one-to-one DM it keeps separate notes for that conversation, stored with the workspace rather than with your Claude account. Anyone in a channel can read or correct its memory by asking. An Owner can view, edit or delete memory files on the Activity page, and the Claude Code changelog for v2.1.296 extends that to members with the Claude Tag Admin permission. Custom instructions set on a channel or workspace outrank memory. During the beta there is no automatic retention period: memory stays until someone deletes it, and it is not in your organization's data exports.

Does Claude Tag memory reach Claude Code, Cursor or Codex?

Not by any path Anthropic documents. Claude Tag stores channel, workspace and DM notes on Anthropic's side with each paired Slack workspace, and its sessions read them. Claude Code on your machine loads its CLAUDE.md files and its own auto memory (memory docs), and Claude Tag memory is not among them. The Claude Tag API, in beta for Enterprise since 23 October 2026, manages five channel and workspace settings and does not read or export memory. It runs the other way too: nothing on your machine reaches a Claude Tag session.

Anthropic draws the line itself: “Keep repository conventions in CLAUDE.md. Put channel conventions in memory.” So a webhook rule your team settles in #eng-payments and Claude Tag saves to that channel's notes is there for the next Slack task, and absent from the engineer who opens the repo in Claude Code, Cursor, Codex or Copilot. To reach them it has to land in a file they read. You can ask Claude Tag in the thread to open a pull request that adds it to CLAUDE.md or AGENTS.md.

Can Claude Tag use skills, plugins and MCP servers?

• Skills and plugins: an Owner or Claude Tag admin attaches them to the whole organization, a workspace or a channel, or syncs them from a skills repository Claude can update. Channel members can add plugins unless an admin restricts it (Customize Claude Tag).
• Repository skills: the skills in a cloned repository's .claude/skills/ load like they do in Claude Code on the web.
• MCP servers: only as connectors an admin adds, including a custom MCP server. Your personal claude.ai connectors can run in a channel for your own requests, after you allow it.

Is Claude Tag free? Plans and pricing

No. Claude Tag is on Team and Enterprise plans only, not Free, Pro or Max, and not on third-party deployments. There is no per-seat charge. Channel, thread and group DM work bills by usage to the organization's usage balance, under a monthly spend limit an Owner sets. A one-to-one DM runs on the sender's own claude.ai seat and its usual limits. The docs give no price per task; to learn what your team's usage costs, Anthropic suggests a pilot with a spend limit set and the per-channel usage page.

Does Claude Tag work in Microsoft Teams?

Not yet. As of 28 October 2026 it works in Slack only, and Anthropic's Teams waitlist page says “Claude Tag launched in Slack, and is now coming to Microsoft Teams,” with no date.

How do I enable Claude Tag in Slack?

• An Owner of your Claude organization opens claude.ai/admin-settings/claude-tag.
• In Slack, send @Claude connect for a pairing code and enter it on the setup page.
• Set a monthly spend limit, add Claude to channels with /invite @Claude, and launch.
• Grant repositories to the Claude GitHub App and add connectors for the tools each channel needs. Claude starts with no access of its own.

Where Claude Tag stops, and how teams fill it

Claude Tag is a strong answer for work a team wants to see and steer in a channel, and its memory is a real team memory, scoped with care. It is a memory for Claude in Slack. It doesn't follow a decision into the terminal, the IDE or another vendor's agent, and a channel note nobody moved into the repo is unknown to every session outside Slack. Teams close that gap three ways:

• Write it into the repo. Rules in CLAUDE.md or AGENTS.md reach Claude Tag and every agent in a checkout. See Claude Code team memory for what else is shared and what stays on one machine, including auto memory.
• Search Slack from the agent. Slack's MCP server lets Claude Code or Cursor look a thread up when someone thinks to ask. Our note on getting Slack decisions into coding agents compares the routes.
• Use a shared memory. A company brain that every agent reads. Our list of company brain tools for coding agents compares them.

Harbor is the last kind: shared memory for AI coding agents. It learns from where your team decides (Slack, pull request reviews, docs and agent sessions), lets you set what waits for a person, and gives Claude Code, Cursor, Codex and GitHub Copilot the rules that apply to each task.