ActionAnthropicUpdated June 2026

How do I use Claude with the computer-use tool?

Short answer: Drop the "AnthropicComputer Use" action anywhere in your workflow, map the inputs from upstream nodes, and publish.

Inputs

The fields this action accepts.

Every field can be mapped from an upstream trigger, AI step, table row, or hard-coded literal.

FieldTypeRequiredDescription
Model
model
optionsRequiredModel. Options: Claude Opus 4.7 (best for agents), Claude Sonnet 4.6
Task
message
stringRequiredOpen the calculator and compute 17 × 23
Display Width
display_width_px
stringOptionalDisplay Width
Display Height
display_height_px
stringOptionalDisplay Height
Max Tokens
max_tokens
stringOptionalMax Tokens
Sample request
{
"model": "{{trigger.model}}",
"message": "Open the calculator and compute 17 × 23",
"display_width_px": "{{trigger.display_width_px}}",
"display_height_px": "{{trigger.display_height_px}}",
"max_tokens": "{{trigger.max_tokens}}"
}
Returns
{
"id": "msg_abc",
"content": [
{
"name": "computer",
"type": "tool_use",
"input": {
"action": "screenshot"
}
}
],
"stop_reason": "tool_use"
}

Use these fields in downstream nodes for routing, logging, or error handling.

Triggered by

Apps that pair well as the trigger for Computer Use.

Any of these apps can fire this action as part of a workflow.

FAQ

Questions about Computer Use.

What does the Computer Use action do in Anthropic?
Sends a message with the computer-use tool, letting Claude take screenshots and click/type on a virtual desktop. Requires Opus 4.x or Sonnet 4.x; you provide the sandboxed desktop environment.
What inputs does Computer Use require?
Required: Model, Task. Every input accepts a static value or a variable from any upstream node in your workflow.
Can I use dynamic inputs from earlier workflow nodes?
Yes. Any field on this action can pull values from upstream nodes, whether that's a form response, a trigger payload, an AI output, or a lookup result.
What happens if Anthropic returns an error?
The workflow pauses on the failed node, the error message is captured in the run log, and you can retry the run with one click. Auto-retry policies are configurable per workflow with exponential backoff up to 5 attempts.
Does Computer Use support batch operations?
Yes. Run Computer Use inside a Loop node to process arrays. Tiny Command handles Anthropic's rate limits automatically so you don't have to throttle manually.
More actions

Other Anthropic actions.

Action
Analyze Image
Sends an image to Claude for analysis (vision), via URL or base64 bytes alongside a text prompt. Use for OCR, chart extraction, accessibility alt-text, or visual QA flows.
Action
Cancel Message Batch
Cancels a message batch that is still processing. Already-completed entries in the batch keep their results; in-flight ones get cancelled with no charge.
Action
Count Tokens
Counts the number of tokens a Messages API call would consume without actually sending the message. Useful for cost estimation and to stay under context-window limits.
Action
Create Message
Sends a message to a Claude model and gets a response. The core Anthropic action; supports system prompts, multi-turn conversation, temperature, and stop sequences.
Action
Create Message Batch
Creates a batch of messages for async processing on the Messages API, with a 50% cost reduction and up to 24-hour turnaround. Perfect for bulk classification, summarization, or labeling jobs.
Action
Create Message with Tools
Sends a message with tool definitions. Claude returns tool_use blocks when it wants to call a function; you execute the call and feed results back as tool_result blocks for the next turn. The foundation for agentic workflows.

Send computer use from your workflows.

Triggered by anything in the catalog. Free tier available. No credit card.