Claude computer use pricing: what it costs on each plan and with the API
What Claude computer use costs on Free, Pro, Max, Team and Enterprise, what the API computer use tool adds in tokens per request and per screenshot, and what ShadowCursor costs on top.
Short answer: Claude's built-in computer use has no price of its own: it is part of the Pro plan ($20 per month, or $17 per month billed annually) and the Max plans ($100 or $200 per month), and it draws on the same usage limits as the rest of Claude and Claude Code. Free, Team and Enterprise plans do not include it. On the API you pay normal token prices: about 4,500 extra input tokens per request for the toolset plus every screenshot as image input. ShadowCursor, a separate computer use MCP server, costs $12.99 per month or $69.99 per year on top of whatever plan or API key your AI runs on.
What each Claude plan costs and includes
Prices from Anthropic's pricing pages as of October 2026, in US dollars and without applicable tax:
| Plan | Price | Built-in computer use |
|---|---|---|
| Free | $0 | No |
| Pro | $20 per month, or $200 per year ($17 per month) | Yes |
| Max 5x | $100 per month | Yes |
| Max 20x | $200 per month | Yes |
| Team, standard seat | $25 per seat per month, or $20 billed annually | No |
| Team, premium seat | $125 per seat per month, or $100 billed annually | No |
| Enterprise | $20 per seat per month billed annually, 20 seats minimum, plus usage at API rates | No |
The Max prices are for web subscriptions. Computer use is a research preview in Claude Desktop and Cowork (macOS and Windows) and in the Claude Code CLI (macOS), and both require Pro or Max. Sources: Claude pricing, Max plan, computer use in Cowork, Claude Code computer use.
How to turn it on once you have the plan is covered in Claude computer use on the Mac.
What is included, and what uses up your limits
Anthropic lists no surcharge and no separate quota for computer use. It runs on your plan's normal usage, which Anthropic describes as shared: "all activity in both tools counts against the same limits", meaning Claude and Claude Code (source). Session limits reset every five hours, and a weekly limit applies across all models (source). When you reach a limit, Pro and Max users can buy usage credits to keep working.
What Anthropic does not publish is how much of your limit a computer use task takes. Our assumption, not an official figure: because Claude looks at a screenshot for almost every step, a long computer use task will use noticeably more than a text conversation of the same length. Keep tasks focused, and let Claude use a connector, the shell or Claude in Chrome where those fit; Anthropic notes that screen interaction is slower than connectors.
Two details that often confuse:
- Max includes API credits of $100 to $200 per month according to the pricing page, but Anthropic's Claude Code article says they cannot be used in Claude Code or the Claude apps, including for extra usage.
- Team and Enterprise cannot buy their way into the built-in feature; it is not available on those plans at any seat type. The routes that do work are in Claude computer use on a Team or Enterprise plan.
What the API computer use tool costs
Developers who build their own agent use the computer use tool of the Claude API. It follows standard tool use pricing, with three parts (source):
- Toolset overhead. Declaring
computer_toolset_20260801adds about 4,500 input tokens to every request (about 4,520 on most models, about 4,590 on Claude Sonnet 5). Turning off zoom saves about 410. - Screenshots. Each screenshot or zoom image is billed as image input. An image costs ⌈width / 28⌉ × ⌈height / 28⌉ visual tokens (source); Anthropic puts typical screenshots at roughly 1,000 to 1,800 input tokens each (source).
- Everything else. Text results, Claude's output tokens and any other tools you use.
Applied to the screen size Anthropic recommends for web apps, 1280 by 800, a screenshot is 46 × 29 = 1,334 tokens. With the published input prices (source), that gives:
| Model | Input / output per million tokens | 1,000 screenshots at 1280 by 800 | Toolset overhead per request |
|---|---|---|---|
| Claude Opus 5.5 | $4 / $20 | about $5.34 | about $0.018 |
| Claude Sonnet 5.5 | $2 / $10 | about $2.67 | about $0.009 |
| Claude Haiku 5.5 (prompts up to 100,000 tokens) | $0.10 / $0.50 | about $0.13 | about $0.0005 |
These are our calculations from Anthropic's rates, not measured task costs. The real bill depends mostly on history: every request resends the conversation, including earlier screenshots, so a long agent loop pays for old images again and again. Anthropic advises keeping screenshot history in check (on the 5.5 models by resizing screenshots and clearing old tool results on the server rather than pruning them in your client), and prompt caching makes repeated parts cheaper (cache hits on Opus 5.5 and Sonnet 5.5 cost 0.05 times the base input price). On Haiku 5.5, prompts over 100,000 tokens cost $0.50 per million input tokens instead.
On top of tokens you pay for the environment, because the API tool does not run anything itself: your application performs the clicks in a virtual machine or container you operate.
What ShadowCursor costs on top
ShadowCursor is a Mac app and MCP server. It is not an AI model and does not replace your Claude plan: the model runs in your client and is billed there. What ShadowCursor adds is the part that operates the apps. Every connected agent gets its own cursor, drawn in its color with a name tag, its own keyboard whose keys arrive only in its window, its own clipboard and its own windows in the background, and several agents can work at once.
| Item | ShadowCursor |
|---|---|
| Trial | 7 days free, no card needed |
| Monthly | $12.99 (EUR 12.99 in Germany), taxes included |
| Yearly | $69.99 (EUR 69.99 in Germany), taxes included |
| Devices | Up to 3 |
| Agents | Unlimited |
| Free tier | None |
Because it is an ordinary MCP server, it also works where Claude's built-in feature is not available: Claude Code on Team or Enterprise or with an API key, and Codex, Cursor or VS Code. It needs macOS 15 or later; a Windows version is in beta. Limits to weigh before you pay: minimized and hidden windows cannot be operated, system drag and drop of files is cancelled, and hover without a click does not reach Chrome or Electron apps.
What a typical setup costs
| Your situation | What you pay |
|---|---|
| Claude Desktop on Pro, one task at a time | Pro only. The built-in computer use is enough; you need nothing else. |
| Claude Code CLI on Max, testing your own app | Max only, as long as one session at a time and hidden apps are fine. |
| Claude Code on a Team or Enterprise seat | Your seat plus a computer use MCP server, for example ShadowCursor at $12.99 per month. |
| Claude Code and Codex working in apps side by side | Each AI's plan plus ShadowCursor once; the license covers unlimited agents. |
| Your own product or pipeline | API tokens as calculated above, plus the VM or container you run. |
Free alternatives exist too: open-source computer use MCP servers cost nothing beyond your model usage. The comparison of computer use MCP servers for the Mac explains what to check before you rely on one.
FAQ
Is computer use included in Claude Pro?
Yes. Pro and Max include computer use in Claude Desktop, Cowork and the Claude Code CLI as a research preview. The free plan, Team and Enterprise do not.
Does Claude computer use cost extra?
Not on Pro or Max: Anthropic lists no surcharge. It uses your plan's shared usage limits, and Anthropic does not say how much a computer use task consumes. On the API you pay for the tokens, including screenshots.
How many tokens does a computer use screenshot cost on the API?
⌈width / 28⌉ × ⌈height / 28⌉ visual tokens; a 1280 by 800 screenshot is 1,334 tokens. Anthropic gives roughly 1,000 to 1,800 tokens per screenshot as a typical range, plus about 4,500 tokens of toolset overhead per request.
Is there a rate limit for Claude computer use?
Anthropic documents no separate limit for computer use. On Pro and Max it counts against the five-hour session limit and the weekly limit shared across Claude and Claude Code.
Does ShadowCursor include Claude?
No. ShadowCursor gives your AI a cursor, keyboard, clipboard and windows of its own; the AI itself runs on your own plan or API key. ShadowCursor costs $12.99 per month or $69.99 per year after a 7-day free trial.