Claude computer use on a Team or Enterprise plan, an API key, Bedrock or Vertex

Short answer: Claude's built-in computer use, in Claude Code and in Claude Desktop, requires a Pro or Max plan. It is not available on Team or Enterprise plans, with an API key or third-party providers such as Amazon Bedrock, Google Cloud or Microsoft Foundry, and not in claude -p. The restriction belongs to that built-in feature, not to MCP: by default Claude Code can connect any MCP server, so a computer use MCP server such as ShadowCursor gives Claude Code computer use on these setups too, unless your administrator restricts MCP servers.

What the built-in feature requires

Anthropic documents the requirements for computer use in the Claude Code CLI plainly. It is a research preview on macOS that requires a Pro or Max plan and is not available on Team or Enterprise plans. You must be authenticated through claude.ai; it is not available with third-party providers like Amazon Bedrock, Google Cloud's Agent Platform (Vertex AI) or Microsoft Foundry. And it needs an interactive session, so it does not run in non-interactive mode with -p (source).

Claude Desktop has the same plan rule: computer use there is a research preview on macOS and Windows that requires Pro or Max and is not available on Team or Enterprise (source).

Setup	Built-in computer use (CLI and Desktop)	An MCP computer use server in Claude Code
Pro or Max	Yes	Yes
Team or Enterprise	No	Yes, unless an admin restricts MCP servers
Anthropic API key	No (needs a claude.ai login)	Yes
Amazon Bedrock, Google Cloud, Microsoft Foundry	No	Yes, unless an admin restricts MCP servers

If you are on Team or Enterprise and run /mcp in Claude Code, the computer-use server simply does not appear. Claude Code's troubleshooting list names the plan and the claude.ai login as reasons.

Why an MCP server is not affected

Claude Code's built-in computer use is itself an MCP server, one that Anthropic ships and gates. Any other MCP server is an ordinary tool connection, and Claude Code's documentation says that by default anyone running Claude Code can connect any MCP server they choose (source). Whether Claude Code reaches the model through a Team seat, an API key or Bedrock, the model calls the server's tools the same way.

So a computer use MCP server gives Claude the same kind of abilities: see an app window, click, type, scroll and pick menu items. What you get depends on the server, not on your plan. For what such servers do and how they differ, see computer use MCP servers for the Mac.

Check your organization's MCP policy first

On managed machines, administrators can restrict MCP servers. Claude Code supports several patterns, documented on its page about controlling MCP server access (source):

A fixed set: a managed-mcp.json deployed by the organization. Users cannot add any other MCP servers.
An approved catalog: allowedMcpServers together with allowManagedMcpServersOnly, so only listed servers load.
A denylist: deniedMcpServers blocks specific servers and allows everything else.
MCP disabled entirely: a managed-mcp.json with an empty server list.

If a policy blocks a server, claude mcp add reports that it is not allowed by enterprise policy, and a server blocked later disappears from /mcp without further notice. If that happens, ask your administrator to approve the server you want to use. Do not try to work around the policy.

Option 1: a separate Pro or Max account

Anthropic's own suggestion for people who reach Claude only through a third-party provider is a separate claude.ai account for this feature (source). That means a second subscription and a second login, and on a company machine it may conflict with your organization's rules on which accounts you may use. Check before you sign up.

Option 2: build your own environment with the API

Developers building their own agent on the Claude API can use Anthropic's computer use tool. That is a different thing: Claude returns actions such as clicks and keystrokes, and your application must run them in an environment you control, typically a virtual machine or container, and send back screenshots. Anthropic recommends a dedicated VM or container with minimal privileges (source). It suits products and automation pipelines, not working with Claude Code on your own Mac.

Option 3: a computer use MCP server on your Mac

For Claude Code on your Mac, a computer use MCP server is the direct route. ShadowCursor is one: a Mac app and MCP server that gives Claude its own cursor, keyboard, clipboard and windows.

Its own cursor, drawn in its color with a name tag.
Its own keyboard. Keys arrive only in the window Claude works in.
Its own clipboard. Copies and pastes never touch your clipboard or another agent's.
Its own windows. Claude works in background windows, even when yours cover them, and screenshots capture only those windows.

Several sessions can work at once, each with its own cursor, unlike the built-in server's one-session lock. Input moves like a person, with curved pointer paths and typing at about 200 words per minute. The same server also works in Codex, Cursor and VS Code, so a team that uses more than one client needs one tool, not several. More on that in running several AI agents at once.

Setup with Team, Enterprise, an API key or Bedrock

Open ShadowCursor and allow Accessibility and Screen Recording. Any computer use tool on a Mac needs these two permissions.
Connect Claude Code with one click. The app finds it, together with Codex, Cursor and VS Code. Nothing depends on how you sign in to Claude.
Start Claude Code as usual and ask for a task in an app.

If your organization uses an MCP allowlist, your administrator has to approve ShadowCursor before step 2 works.

What to tell your security team

ShadowCursor runs locally on the Mac. Input goes to one app and one window only.
Screenshots capture only the window the agent works in, even when it is covered. They go to the model your Claude Code uses, as with any computer use.
An agent cannot see minimized or hidden windows.
ShadowCursor needs macOS 15 Sequoia or later, and the same two macOS permissions as Anthropic's built-in computer use.

Limits to know

Minimized and hidden windows cannot be seen or operated.
System drag and drop of files, tabs or text follows the real mouse, so ShadowCursor cancels such drags.
Hover without a pressed button works in Safari, not in Chrome or Electron apps.
Code editors such as VS Code do not expose their text through accessibility; agents read them from the screenshot.

FAQ

Is Claude computer use available on the Team plan?

Not the built-in feature. Anthropic's documentation says computer use in Claude Code and Claude Desktop requires Pro or Max and is not available on Team or Enterprise. Claude Code on a Team plan can still use a computer use MCP server.

Does computer use work with Amazon Bedrock or Google Vertex AI?

The built-in Claude Code feature does not; it needs a claude.ai login. An MCP server works with Claude Code regardless of the provider. Developers can also use the computer use tool of the Claude API, with an environment they run themselves.

Can I use computer use with an API key?

Not the built-in feature in Claude Code, which requires a claude.ai login on Pro or Max. With an API key, Claude Code can use a computer use MCP server such as ShadowCursor.

Can my admin block computer use MCP servers?

Yes. Administrators can deploy a fixed set of MCP servers, allow only approved ones, block specific servers or turn MCP off entirely through managed settings.

What does ShadowCursor cost?

A 7-day free trial without a card, then $12.99 per month or $69.99 per year, with unlimited agents on up to three devices. It is a separate subscription from your Claude plan.