Claude SDKs add built-in computer and browser use loops
- Source
- x.com
- Date
Building a computer-use agent used to mean writing the loop yourself: read Claude's requested action, translate each click or keystroke into a driver command, capture the result and send it back. The SDKs now ship that loop as built-in toolsets for both computer use and browser use, so developers plug in a driver instead of maintaining the glue code. That removes a common source of bugs in early agents, where action schemas drift from the driver's API. The timing lines up with Haiku 5.5, which Anthropic markets for computer use at a much lower price, so the cost and the boilerplate of running a screen-driving agent both dropped in the same week. Teams with hand-written loops should compare them against the SDK version before adding features.
Computer use and browser use toolsets are now built into the Python and TypeScript SDKs for Claude. The API tells you what Claude wants to click or type. Previously, you had to write your own loop and map clicks and keystrokes to commands, but the SDKs now run the loop and send actions to drivers.
Context
Anthropic has moved the repetitive part of building a computer-using Claude out of developers' code and into its own libraries. Computer use lets Claude operate a screen or a web browser, but the only answers with what Claude wants to do next, such as a click or a piece of typing. As ClaudeDevs, the company's developer account, explains, developers previously had to write the loop themselves: take each requested click or keystroke, translate it into a command that something could actually execute, and repeat until the task was done.
The Python and TypeScript SDKs, the libraries developers use to call Claude from those languages, now include computer use and browser use toolsets that run that loop and pass each action to a driver, the component that carries it out on a real browser or machine. The capability itself is not new: computer use and the browser became generally available on the Claude Platform in August. What changes is how much glue code a developer writes to use them. The post does not say which drivers are supported or how they are set up, and the attached demo video could not be read, so those details are not covered here.
- Agent — A system in which a model repeatedly chooses operations, a harness carries them out, and the results return to the model to inform its next choice.
- API — Application programming interface: the agreed set of operations, inputs and outputs through which one piece of software uses another.
- Tool — An operation that the software around a model makes available for the model to request, with a name, expected inputs and a kind of result.
Checking sign-in…
Loading comments…






