Cursor computer use on the Mac: local apps instead of a cloud VM

Short answer: Cursor's own computer use runs in Cloud Agents: each gets an isolated VM with a desktop and browser that it operates with mouse and keyboard. You can also start a self-hosted worker with agent worker --computer-use start to let a cloud-run agent drive a Mac's signed-in desktop session. The agent in the Cursor editor itself has a built-in Browser tool for web pages; Cursor documents no editor tool for native Mac apps. To let that local agent operate apps on your Mac, connect a computer use MCP server; Cursor calls MCP tools automatically and passes returned screenshots to the model.

What Cursor's own features cover

Cursor offers three different things that touch on computer use. They differ in where the agent runs and what it can reach:

Sources: Cloud Agent capabilities, computer use and desktop sharing, My Machines, Browser.

Cloud Agents

Each cloud agent runs in its own isolated VM with a full desktop. It can start a dev server, open the app in a browser, click through flows and check its work before it opens a pull request, then attach screenshots, videos and logs as artifacts. You can take over the VM's remote desktop and hand control back. This is a strong fit for web apps and anything that runs in a reproducible environment, because nothing happens on your own Mac.

Self-hosted workers

A worker connects your machine to Cursor. On a Mac, starting it with --computer-use installs a helper app called Cursor Computer Use, which needs Accessibility and Screen Recording of its own. Cursor's documentation says the worker "drives the signed-in desktop session" of the Mac user that runs it; it does not describe background windows, a separate cursor or several agents sharing one desktop. Desktop sharing with an isolated agent desktop is Linux only. Tasks for a worker are sent from cursor.com/agents, not from the chat in your editor.

The Browser tool

The agent in the editor can navigate, click, type, scroll, take screenshots and read console output and network traffic in Cursor's browser, without installing external tools (on Enterprise teams an admin turns browser features on first). Cookies and storage persist between sessions, isolated per workspace. For testing a web app you are building, it is often all you need.

When you need a local computer use server

None of the three covers the case many Mac developers have: the agent in the Cursor editor or the Cursor CLI, working on your Mac, needs a native app. Typical examples:

• Launching the macOS or iOS simulator build it just compiled and clicking through the new screen.
• Checking that an exported file opens correctly in Numbers, Preview or another desktop app.
• Working through a file dialog, a settings window or a menu that no API reaches.
• Testing a website in Safari, in a real browser window rather than an embedded one.

A VM cannot run the apps on your Mac, and a worker is a different workflow from the chat you are in. A computer use MCP server brings these tools into the editor agent itself.

How MCP tools work in Cursor's agent

Cursor's MCP documentation describes the parts that matter for a computer use server:

• Installation. Servers are installed from the Customize page, from the marketplace, or through mcp.json: ~/.cursor/mcp.json for all projects, .cursor/mcp.json for one project. Local servers run over stdio; remote servers use SSE or Streamable HTTP.
• Automatic use. The agent uses MCP tools listed under Available Tools whenever they are relevant, including in Plan Mode. You can also name a tool or server in your prompt.
• Approval. By default Cursor asks before it runs an MCP tool and lets you inspect the arguments. MCP follows the same Run Modes as terminal commands; in Auto-review, allowlisted tools run directly and the rest go through a classifier.
• Screenshots. MCP servers can return images, and Cursor attaches them to the chat. If the model supports images, it analyzes them. That is what makes computer use work: the agent sees the window it acts in.
• On and off. Each server has a toggle in Customize. A disabled server does not load.
• The CLI too. The Cursor CLI detects and respects the same mcp.json, so servers you set up for the editor also work in the terminal (source).
• Team policy. Enterprise admins can restrict which MCP servers and tools a team may run.

ShadowCursor: computer use for Cursor's agent on your Mac

ShadowCursor is a Mac app and MCP server. When Cursor connects, its agent gets a full set of its own:

• Its own cursor, drawn on screen in the agent's color with a name tag.
• Its own keyboard. Keys arrive only in the agent's window, in one app.
• Its own clipboard. Copy and paste never touch your clipboard or another agent's.
• Its own windows. The agent works in background windows, even when yours cover them, and screenshots capture only the window it works in.

Every action returns the window's new state, a screenshot plus accessibility elements, so the agent sees the effect of each click. Several agents can work at once, each with its own cursor: Cursor on one project, Claude Code or Codex on another. Input moves like a person, with curved pointer paths and typing at about 200 words per minute, so hover menus in Safari, sliders and date pickers react as they do to you. It runs locally and needs macOS 15 Sequoia or later.

Setup in Cursor

1. Open ShadowCursor and allow Accessibility and Screen Recording when your Mac asks.
2. Connect Cursor with one click. The app finds Cursor on your Mac and registers itself. There is no mcp.json to edit.
3. Check Customize. ShadowCursor appears with its tools; make sure its toggle is on.
4. Ask the agent for an app task. For example: "Run the app, open Preferences, switch the theme to dark and check that the sidebar follows." Approve the first tool calls, or allowlist the tools in your Run Mode once you trust the workflow.

Cloud VM or local apps: which one

Limits to know

• Minimized and hidden windows cannot be seen or operated. Covered windows can.
• System drag and drop of files, tabs or text follows the real mouse, so ShadowCursor cancels such drags instead of dropping.
• Hover without a pressed button works in Safari, not in Chrome or Electron apps.
• An agent cannot use a background tab of your browser window; it opens its own window.
• Code editors such as VS Code do not expose their text through accessibility, so agents read them from the screenshot.
• ShadowCursor is a separate subscription from Cursor: a 7-day free trial without a card, then $12.99 per month or $69.99 per year.

FAQ

Does Cursor have computer use?

Yes, in Cloud Agents, which run in isolated VMs with a desktop and browser, and on self-hosted workers started with --computer-use. The agent in the editor has a Browser tool for web pages. For native apps on your Mac, it needs a computer use MCP server.

Can Cursor's agent control my Mac?

With a self-hosted worker, a cloud-run agent can drive your Mac's signed-in desktop through Cursor's helper app. With an MCP server such as ShadowCursor, the agent in your editor or CLI operates Mac apps in windows of its own, with its own cursor, keyboard and clipboard.

Does Cursor computer use work on Windows?

Cloud Agents run in Cursor's VMs, so your own operating system does not matter. Cursor documents self-hosted computer use for macOS and Linux workers. ShadowCursor's Windows version is in beta.

Why does the agent not use my MCP server?

Check that the server's toggle is on in Customize and that its tools appear under Available Tools. Then name the server in your prompt. If your team uses an MCP allowlist, an admin may have to approve it.

Can Cursor and Claude Code operate apps at the same time?

Yes, with ShadowCursor. Each agent connects on its own and gets its own cursor, keyboard and clipboard. Give each one its own window or document. See running several AI agents at once.