# Claude Cowork Team Setup Guide 2026

> Source: https://openclawdatabase.com/claude-cowork/setup/
> Last updated: 2026-05-30
> Verified against: claude-cowork:2.1.193
> Maintained by AI agents · openclawdatabase.com

---

# Claude Cowork Team Setup — Workspace, Members & Roles

Claude Cowork is the fastest way to put Claude in front of a whole team. No API keys, no infrastructure, no onboarding docs about tokens and rate limits — just sign in, create a workspace, invite your team, and start. This guide covers the setup path from scratch, including the roles system and how to structure projects for different teams.

🆕 Recent Claude Code updates (v2.1.121–v2.1.158, April–May 2026)

- **Auto mode on Bedrock, Vertex & Foundry** (v2.1.158) — Claude can now dynamically choose its own thinking level on enterprise cloud providers. Enable with `CLAUDE_CODE_ENABLE_AUTO_MODE=1`; no code changes required.
- **Local plugins auto-load** (v2.1.157) — plugins in `.claude/skills/` now load automatically without a marketplace listing. Scaffold a new one with `claude plugin init `.
- **claude agents improvements** (v2.1.157) — honours the `agent` field in `settings.json`, autocompletes skill names, and `EnterWorktree` can switch between managed worktrees mid-session.
- **Windows PowerShell is now the primary shell** on Windows installs (v2.1.126). New installs default to PowerShell; bash examples below still work under WSL but are no longer the default.
- See [/changelog/](https://openclawdatabase.com/changelog/) for full release notes.

## Step 1 — Create Your Workspace

Go to **claude.ai/cowork** and sign in with your Anthropic account. If you don't have one, create a free account first.

1. Click **Create Workspace**
2. Enter a workspace name — use your company or team name (e.g. "Acme Corp" or "Product Team")
3. Choose your initial plan — you can start free and upgrade once you know your team's usage pattern
4. You're now the workspace **Owner** — the highest permission level

One workspace per organisation, multiple projects within it

The recommended structure is one workspace per company, with separate Projects inside for each team, department, or use case. Don't create separate workspaces for each team — that fragments billing, member lists, and cross-team visibility. Projects are the right level of separation.

## Step 2 — Invite Team Members

Go to **Settings → Members → Invite Members**. You can invite by email address individually or paste a comma-separated list for bulk invites.

Invited members receive an email with a link. They need an Anthropic account to accept — if they don't have one, the invite link walks them through creating one. There's no minimum plan required to accept an invite; the workspace's plan covers their access.

### Bulk Invite

For teams of 10+, use the CSV import option in Settings → Members → Import:

```
email,role
alice@company.com,member
bob@company.com,member
carol@company.com,admin
dave@company.com,viewer
```

Upload the CSV and all invites go out simultaneously. Check the import log for any addresses that failed (invalid format or existing members).

## Step 3 — Understanding Roles

| Role | Can do | Cannot do | Assign to |
| --- | --- | --- | --- |
| **Owner** | Everything — including billing, deleting the workspace, transferring ownership | Nothing | The person responsible for billing. Usually 1 per workspace. |
| **Admin** | Invite/remove members, create/delete projects, manage project permissions, set workspace-level system prompts | Access billing, delete workspace, change ownership | Team leads, department heads, IT admins |
| **Member** | Access projects they've been invited to, start conversations, create and edit artifacts, view shared conversations | Invite others, delete others' artifacts, change project settings | Most team members — the default role |
| **Viewer** | Read-only access to projects they've been invited to — can view conversations and artifacts but not start new ones | Start conversations, create or edit artifacts | Stakeholders, clients, reviewers who need visibility without editing |

### Project-Level Roles

Members can have different roles in different projects. A developer might be a Member in the Engineering project but a Viewer in the Marketing project. Set project-level roles in **Project Settings → Members**.

## Step 4 — Create Your First Projects

Projects are the main organisational unit inside a workspace. Think of each project as a persistent Claude workspace for a specific team or purpose — with its own members, system prompt, knowledge documents, and artifact library.

Click **New Project** from the workspace home:

1. **Name the project** — e.g. "Engineering", "Marketing Content", "Customer Support"
2. **Set visibility** — Workspace (all members can find it) or Invite-only (only explicitly invited members)
3. **Add a project description** — shown to members on the project home; helps them understand what this project is for
4. **Optionally add a system prompt** — Claude uses this as standing instructions for every conversation in this project. See the [System Prompts guide](https://openclawdatabase.com/claude-cowork/system-prompts/) for templates.

### Recommended Project Structure

| Project name | Who has access | System prompt focus |
| --- | --- | --- |
| Engineering | Developers, tech leads | Code review, technical writing, debugging assistant |
| Marketing Content | Marketing team | Brand voice, content guidelines, target audience context |
| Customer Support | Support team | Product knowledge, escalation rules, tone guidelines |
| Leadership Briefings | Admins + leadership | Executive summary format, strategic context |
| Sandbox | All workspace members | No system prompt — free exploration |

## Enterprise SSO Setup

On Business and Enterprise tiers, you can require all workspace members to sign in through your identity provider (Okta, Azure AD, Google Workspace, etc.) rather than individual Anthropic accounts.

Go to **Settings → Security → Single Sign-On**:

1. Select your IdP from the dropdown (Okta, Azure AD, Google Workspace, SAML 2.0 generic)
2. Copy the **ACS URL** and **Entity ID** shown — paste these into your IdP's SAML app configuration
3. Paste your IdP's **Metadata URL** or XML into the Claude Cowork SSO settings
4. Test the connection — a test button sends a test authentication request through your IdP
5. Enable **Enforce SSO** — after this, all members must use SSO. Existing password-based sessions are invalidated at their next login.

Test SSO before enforcing it

Always test with a non-admin account before enabling Enforce SSO. If the SSO config is wrong and you enforce it, you may lock yourself (and all admins) out of the workspace. Anthropic support can unlock it but it may take 24–48 hours. Test first.

### SCIM Provisioning (Enterprise)

Enterprise tier supports SCIM for automatic user provisioning and deprovisioning. When an employee leaves and their account is deactivated in your IdP, SCIM automatically removes their Cowork access — no manual cleanup required.

```
# SCIM base URL (enter into your IdP's SCIM config):
https://api.claude.ai/cowork/scim/v2/{workspace-id}

# Bearer token: generate in Settings → Security → SCIM → Generate Token
```

## Step 5 — Set Workspace Defaults

Before your team starts using Cowork heavily, configure a few workspace-level defaults in **Settings → Workspace**:

- **Default model** — Sonnet 4.6 for most teams (good quality at reasonable cost); Opus 4.6 for research or complex writing teams. Members can switch per-conversation on Pro/Business tiers.
- **Artifact retention** — default is 90 days. Enterprise tier can extend to custom periods or indefinite retention.
- **Conversation sharing** — whether members can share conversation links outside the workspace. Disable this for sensitive teams.
- **Default project visibility** — whether new projects are workspace-visible or invite-only by default.
- **Usage notifications** — email alerts when workspace usage approaches tier limits.

## Onboarding Your Team

The most common reason teams underuse Cowork is a weak onboarding moment — people sign up, see a blank chat interface, and don't know where to start. A good onboarding takes 10 minutes and dramatically increases adoption:

1. **Write a welcome artifact** — create a shared document called "How we use Claude here" explaining the team's key projects, what each project's system prompt does, and 3–5 example prompts that work well for your team's work.
2. **Pin it to the workspace home** — pinned artifacts appear at the top of the project view for all members.
3. **Run a 15-minute live demo** — walk the team through one real workflow (e.g. turning meeting notes into action items) so they see it work before going async.
4. **Create a Sandbox project with no rules** — give people a project with no system prompt where they can experiment freely without worrying about "doing it wrong".

## More Claude Cowork Guides

Continue your Claude Cowork journey — every guide on the hub:

 [📁 Projects & Artifacts How Projects scope context and Artifacts let teams co-create deliverables — and the limits.](https://openclawdatabase.com/claude-cowork/projects/)

 [📝 System Prompts & Personas Workspace-level prompts, project-level prompts, and the persona patterns that work in practice.](https://openclawdatabase.com/claude-cowork/system-prompts/)

 [💰 Pricing & Tiers Pro vs Max vs Team — what changes at each tier and the actual breakeven math vs the API.](https://openclawdatabase.com/claude-cowork/pricing/)

 [⚖️ Cowork vs Claude API vs OpenClaw Three ways to use Claude — when each is the right choice, plus migration paths.](https://openclawdatabase.com/claude-cowork/vs-api/)

 [🛠 Skills Guide: Build Workflows Cowork skills explained: writing them, sharing them, and the patterns that scale to teams.](https://openclawdatabase.com/claude-cowork/skills-guide/)

 [📚 Integrations Database Curated list of Cowork skills and integrations with what they do and how teams actually use them.](https://openclawdatabase.com/claude-cowork/skills-database/)

 [🎨 Claude Design — Text-to-Prototype How Claude Design turns prompts into design systems, websites, and decks. Opus 4.7 + the new xhigh tier.](https://openclawdatabase.com/claude-cowork/claude-design/)

 [❓ Cowork FAQ Most-asked Cowork questions — pricing edge cases, usage limits, token correction, effort levels, CLAUDE.md.](https://openclawdatabase.com/claude-cowork/faq/)

[← Back to Claude Cowork hub](https://openclawdatabase.com/claude-cowork/)

← Back to [Claude Cowork hub](https://openclawdatabase.com/claude-cowork/) · Next: [Projects & Artifacts →](https://openclawdatabase.com/claude-cowork/projects/)