Back to Guides
/ Guides

Claude in Slack: How Claude Tag Works (and How to Make It Build Things)

Aleks Koha7 min read

In June 2026 Anthropic shipped Claude Tag, and the pitch is simple: instead of Claude being a window you open, it's a teammate in your Slack workspace. Anyone in a channel types @Claude, hands it a task, and gets back to their own work while Claude posts threaded updates.

It's a genuinely new shape for AI at work — multiplayer, persistent, proactive. It's also, out of the box, a teammate with a particular limitation, and the fix for that limitation is the more interesting story. Both halves below.

What Claude Tag Is

Claude Tag is Claude as a member of your Slack workspace — with its own identity, its own memory, and access to whatever tools, data, and codebases your admin grants it. Invite it to a channel with /invite @Claude, and from then on anyone there can delegate work to it.

Anthropic frames it as the evolution of Claude Code into something team-shaped, and dogfoods it hard: 65% of their product team's code is reportedly created by their internal version. Usage has spread beyond engineering — chasing down metrics, working support tickets, root-causing bugs.

It runs on Opus 4.8 and is in beta for Claude Enterprise and Team plans.

How @Claude Works in a Channel

Four properties make it different from a chat window:

It's multiplayer. One Claude per channel, shared by everyone. Anyone can see what it's working on and pick up a thread where a colleague left off. It behaves like a teammate, not like everyone's private assistant.

It learns over time. Claude follows the channel and accumulates context, so you stop re-explaining the project every time. With permission, it can also learn from other channels and data sources (private channels stay private).

It takes initiative. With "ambient" behavior enabled, Claude flags things it thinks you should know and follows up on threads that went quiet without resolution.

It works asynchronously. Give it a task and go do something else; it breaks the task into stages and posts results in the thread. It can even schedule tasks for itself and pursue a project over days — the same idea we covered in Claude Scheduled Tasks, embedded in a channel.

Admins control everything that matters: which tools and data @Claude can touch per channel, token-spend limits per org and per channel, and a full log of what Claude did and who asked.

Who Gets It (and What It Replaces)

Claude Tag is in beta for Claude Enterprise and Team customers, and it replaces the older "Claude in Slack" app (admins get a migration window). Anthropic has said the goal is to expand availability — but today, if you're not on those plans, you're not tagging @Claude. (There's a route for you anyway; see below.)

Setup, for those who have it: pair Claude Tag with your Slack workspace from the Claude admin settings, grant tool access, set a spend limit, and test in a private channel before letting it loose.

The Gap: A Teammate That Can't Ship

Here's the limitation. @Claude's abilities are bounded by the tools your admin connects — and the default toolset is oriented around knowledge work: reading your docs, querying your data, writing code and text. Ask @Claude to analyze churn and it delivers. Ask it to stand up the churn dashboard — an actual page, with a database behind it, refreshed nightly — and it hits the same wall every coding agent hits: writing the thing is not shipping the thing. Someone still has to provision servers, wire the database, manage credentials, and deploy.

So the team's Slack-native AI produces artifacts that a human then carries the last mile. The bottleneck moved; it didn't disappear.

Building From Slack: @Claude + Matagi MCP

Claude Tag's tool access is the hook: connect the Matagi MCP server as one of @Claude's tools, and the last mile becomes something Claude does itself. Matagi exposes infrastructure as MCP tools — provision serverless functions and isolated Postgres databases, deploy, register cron schedules, run batch jobs, and stand up always-on agents — all behind one remote endpoint with OAuth, no API keys.

That changes what a Slack message can accomplish:

@Claude build a status page for our API, host it, and have it check the endpoints every five minutes. Post here if anything goes red.

With Matagi in its toolbox, @Claude can write the checker, provision the database for uptime history, deploy the page, register the schedule, and reply in-thread with a link to the running thing. The team watched it happen in the channel; the audit trail exists on both sides (Slack thread + Matagi's action log); and credentials never appeared anywhere in the conversation — Matagi's proxy injects them server-side.

The mental model: Claude Tag gives your team a shared brain in Slack; Matagi gives that brain hands. One is where work is discussed and delegated; the other is where the built things actually live and run. Because Matagi's endpoint is client-agnostic, the same workspace is reachable from Claude Code, Cursor, ChatGPT, or Codex — so what gets built from Slack isn't trapped in Slack.

The usual rule applies: let the first few builds deliver to a test channel or a staging URL, review, then trust. Automate the volume, keep the judgment.

Not on Enterprise? The Other Route Into Slack

If your team isn't on Claude Enterprise or Team, you can still get an AI teammate in Slack — from the other direction. Matagi agents are always-on workers that wire directly into Slack (also Telegram, Discord, email, and web chat). You describe the agent in plain language — what it watches, what it does, where it posts — connect it to a channel, and it operates there like Claude Tag's independent cousin: answering, monitoring, and, because it lives on Matagi, building and running infrastructure natively.

We covered the general pattern in How to Build a Slack Bot With No Code — the difference in 2026 is that the "bot" can now be a full agent with provisioning capability, not a canned-response script.

FAQs

What is Claude Tag in one sentence? Claude as a persistent, multiplayer teammate in your Slack channels — tag @Claude, delegate a task, and it works asynchronously with memory and admin-scoped tool access.

Which plans have Claude Tag? Beta for Claude Enterprise and Team plans. It replaces the earlier Claude in Slack app.

Can @Claude in Slack deploy real infrastructure? Not by itself — its abilities depend on connected tools. Connect the Matagi MCP server and @Claude can provision databases, deploy apps and endpoints, and register scheduled jobs directly from a Slack conversation.

Is it safe to give a Slack AI provisioning powers? Scope it like you'd scope a new hire. Matagi isolates resources per workspace, keeps credentials in an encrypted server-side proxy (never in the model context or the thread), and logs every action. Combined with Claude Tag's own admin controls and audit log, you get more oversight than most human-run deploy processes have.

Can I DM Claude instead of using channels? Yes — @Claude responds privately in DMs, using the personal tools and connectors you've set up.

Give your team's @Claude hands — connect the Matagi MCP server as one of its tools, or spin up a Matagi agent in your Slack directly. New workspaces start with trial credit.

claude slackclaude tagslack aiai agentsmcp

Build your first AI agent free

Describe what you want done in plain English. Matagi provisions the infrastructure, wires the integrations, and deploys it.

Get started