Claude Tag turns Slack into a workspace where AI participates as a persistent team member. Setting it up properly matters: you want Claude in the right channels, connected to the right tools, with the right permissions. Get it wrong and you either have a useless bot or an AI with more access than it should have.
This guide walks through the complete setup process for Claude Enterprise and Team admins.
Prerequisites
Before you start:
- Claude Enterprise or Claude Team subscription (Claude Tag isn’t available on individual plans)
- Slack workspace admin access (you need to install apps)
- Anthropic Console access (admin-level account at console.anthropic.com)
If you’re not sure which Claude plan you’re on, check your billing at console.anthropic.com. Claude Tag only appears for Enterprise and Team customers.
Step 1: Enable Claude Tag in Anthropic Console
- Log into console.anthropic.com with your admin account
- Navigate to Organization Settings > Integrations
- Find Claude Tag for Slack and click Enable
- Review the permissions summary (what Claude Tag will access in Slack)
- Click Confirm and Continue
This generates the connection between your Anthropic org and your Slack workspace but doesn’t install anything in Slack yet.
Step 2: Install Claude Tag in Slack
- From the Anthropic Console, click Connect to Slack (you’ll be redirected)
- Choose the Slack workspace to connect
- Review the OAuth permissions Slack requests:
- View messages in channels Claude is added to
- Post messages and replies
- Access files shared in those channels
- View channel membership
- Click Allow
After authorization, Claude Tag appears as an app in your Slack workspace. It’s installed but not yet active in any channels.
Step 3: Add Claude to Channels
Claude Tag only operates in channels where it’s been explicitly added. It has zero access to anything else.
Adding to a channel:
- Open the channel in Slack
- Click the channel name at the top to open settings
- Click Integrations > Add Apps
- Search for Claude and click Add
Or use the slash command:
/invite @ClaudeWhich channels to start with:
Start small. Don’t add Claude to every channel on day one. Recommended starting points:
- One engineering channel: Where technical discussions happen
- One project channel: An active project with ongoing decisions
- A test/sandbox channel: For experimenting without stakes
Avoid adding Claude to:
- HR or people-ops channels (sensitive personnel matters)
- Executive/leadership channels (confidential strategy)
- Social/casual channels (people should have spaces without AI observation)
Step 4: Connect Tools and Data Sources
This is where Claude Tag becomes powerful. Without connections, it’s just a chatbot that reads messages. With connections, it can pull real data and take real actions.
Available connection types:
Code repositories:
- GitHub (repos, PRs, issues)
- GitLab (repos, MRs, issues)
- Bitbucket
Documentation:
- Notion
- Confluence
- Internal wikis via URL
Data and APIs:
- Custom API endpoints (REST)
- Database read-only connections
- Internal tools via webhook
Development tools:
- CI/CD pipelines (GitHub Actions, CircleCI, Jenkins)
- Monitoring (Datadog, PagerDuty)
- Project management (Linear, Jira)
Configuring connections:
- In Anthropic Console, go to Claude Tag > Connections
- Click Add Connection
- Select the service type
- Authenticate (OAuth for supported services, API key for custom)
- Configure scope (which repos, which databases, read-only vs read-write)
- Set which channels can use this connection
Important: Each connection can be scoped to specific channels. Your engineering channel might have GitHub access, but your marketing channel doesn’t need it.
Step 5: Configure Permissions
Organization-level permissions:
In Anthropic Console under Claude Tag > Permissions:
| Permission | Description | Default |
|---|---|---|
| Message reading | Can Claude read channel messages | On (required) |
| File access | Can Claude view shared files | On |
| Proactive messages | Can Claude message without being tagged | Off |
| External data access | Can Claude use connected tools | On |
| Code execution | Can Claude run code via Claude Code | Off |
| Write actions | Can Claude make changes (PRs, tickets) | Off |
Recommendations:
- Keep proactive messages off initially. Teams find it jarring when AI speaks unprompted.
- Start with code execution off until the team is comfortable with Claude’s judgment.
- Write actions should be off until you’ve validated Claude’s understanding of your systems.
Channel-level overrides:
You can set per-channel permissions that override org defaults. For example:
- Engineering channel: code execution ON, write actions ON (for experienced devs)
- Marketing channel: code execution OFF (no need)
- Security channel: external data access limited to specific tools
Step 6: Verify the Setup
Test that everything works:
- Go to a channel where Claude is added
- Tag @Claude with a simple question:
@Claude what channels do you have access to? - Test tool access:
@Claude what's the latest commit on our main branch?(if GitHub is connected) - Test context: Have a conversation, then later ask
@Claude summarize what we discussed about [topic]
If something doesn’t work:
- Check that Claude is actually added to the channel (not just installed in the workspace)
- Verify tool connections are authorized and scoped to that channel
- Check Anthropic Console for connection errors
Best Practices for Channel Organization
After watching teams roll out Claude Tag, here’s what works:
Create dedicated project channels
Rather than adding Claude to existing general channels, consider creating project-specific channels where Claude participates. This keeps context focused and makes Claude’s persistent memory more useful.
Example structure:
#project-alpha(Claude added, team discussions about the project)#project-alpha-social(no Claude, team bonding and casual chat)
Use thread discipline
Claude Tag works best when discussions happen in threads. It can follow thread context much better than scattered top-level messages. Encourage your team to reply in threads rather than the main channel.
Set clear norms with your team
Before rolling out, communicate:
- Which channels have Claude (and which don’t)
- What Claude can and can’t do
- When to tag Claude vs. handle something yourself
- That Claude reads all messages in its channels (not just tagged ones)
Regular access reviews
Monthly, check:
- Is Claude in channels it shouldn’t be?
- Are tool connections still appropriate?
- Has the team’s use pattern revealed permission gaps?
- Should Claude be added to new channels?
Onboarding Your Team
What to tell employees:
- What Claude Tag is: A persistent AI teammate in specific Slack channels
- How to use it: Tag @Claude followed by your request
- What it sees: All messages in channels it’s been added to
- What it can do: Answer questions, find information, complete tasks (based on permissions)
- What it can’t do: Access channels it’s not in, perform unauthorized actions
Common first interactions to suggest:
@Claude summarize the decisions we made this week@Claude what do you know about [project]?@Claude find the documentation for [feature]@Claude help me draft a message about [topic]@Claude what's the status of [task] based on recent discussions?
Troubleshooting
| Issue | Solution |
|---|---|
| Claude doesn’t respond | Check it’s added to the channel, not just the workspace |
| ”I don’t have access to that” | Check tool connections and channel-level permissions |
| Slow responses | Normal for complex queries. Async tasks take longer. |
| Inaccurate context | Claude may not have enough historical context yet. Give it time. |
| Too chatty | Turn off proactive messages in permissions |
| Doesn’t remember last week | Context window has limits. Very old discussions may be summarized rather than recalled verbatim. |
What Happens Next
After setup, Claude Tag becomes more useful over time as it accumulates context. The first few days, it’s essentially a smart chatbot. After a week or two, it starts connecting dots between conversations and providing genuinely contextual responses.
For a broader overview of what Claude Tag is and how it compares to alternatives, see our What Is Claude Tag explainer and the comparison with ChatGPT in Slack and Copilot.
If your team also works with AI-powered document processing or vision tasks, you might want to explore how these capabilities integrate with your development workflows. Our DeepSeek Vision complete guide and multimodal AI APIs price comparison cover the technical side of building with AI APIs.
FAQ
How long does setup take?
The basic setup (install, add to channels) takes 10-15 minutes. Connecting tools and configuring permissions adds another 30-60 minutes depending on how many integrations you need. Total: under 2 hours for a complete setup.
Can non-admins add Claude to channels?
By default, only workspace admins can install the Claude Tag app. Once installed, channel-specific addition depends on your Slack workspace settings and Anthropic Console permissions. Most orgs restrict channel addition to admins or channel owners.
What happens if I remove Claude from a channel?
Claude loses access to that channel immediately. It can no longer read messages or respond there. Context it previously gathered from that channel may be retained in its memory for other channels where it’s still active (depending on your data retention settings).
Can I limit Claude to read-only in some channels?
Yes. In Anthropic Console, you can set channel-level permissions. Set a channel to “observe only” and Claude will read messages for context but won’t respond unless explicitly configured otherwise.
How do I handle sensitive information accidentally shared with Claude?
Contact Anthropic support for data deletion requests. You can also remove Claude from the channel immediately to prevent further access. For proactive protection, don’t add Claude to channels where sensitive information is regularly discussed.
Does Claude Tag work with Slack Connect (shared channels)?
This depends on your organization’s security settings. External shared channels may have different permission requirements. Check with Anthropic support for your specific configuration.