Skip to main content
Back to timeline
anthropics/claude-quickstartsSource publication:

Anthropic opens Claude Quickstarts: starter code for desktop, browser and chat agents now sits in one repository

Synopsis

Anthropic's claude-quickstarts repository gathers a set of starter projects built on the Claude API, spanning a customer support agent, financial data analysis, desktop and browser control, an autonomous coding agent, and examples that connect a managed-agent session to a chat surface. Each project supplies a foundation developers can extend, and the minimal browser and computer toolset examples are explicitly described as showing the interface rather than being production code. The repository also names the computer_toolset_20260801 and browser_toolset_20260801 toolset shapes, in which each computer action is a member tool the model calls by name.

Source-provided article image: anthropics / claude-quickstarts
GitHub Trending

Interpretation

Developers can now draw Claude starter projects from one repository by use, covering customer support, financial data analysis, desktop control, browser automation, autonomous coding and chat-surface integration. The repository lists these directions one by one, giving each project a purpose description and an entry link. The customer support agent project is described as using Claude's natural language understanding and generation with access to a knowledge base.

Desktop control comes in two forms: a containerized Computer Use Demo and a native reference implementation that runs directly against the macOS desktop, the latter demonstrating explicit tool definitions, image sizing and pruning, prompt caching, server-side compaction, batched tool calls, a sandboxed shell and trajectory recording. Unlike the containerized demo, the native macOS version runs directly against the macOS desktop and separately lays out common patterns for building more reliable and cost-effective computer-use agents. The native implementation is described as pedagogical and is paired with Anthropic's computer-use best-practices guide, with a note to run it in a VM.

The minimal browser and computer toolset examples are given in Python and TypeScript, each with a small driver, a script that runs the driver with the tool runner, and a script that exercises it without a model. These two examples isolate the toolset interface so developers can verify the driver and tool-call path before involving a model. The repository states plainly that these examples show the interface and are not production code.

Perspective

The projects target developers who need a fast start and provide foundations meant to be extended; the minimal browser and computer toolset examples are explicitly labeled as showing the interface rather than production code, and the native macOS computer-use implementation is described as pedagogical with a recommendation to run it in a VM.

The repository's descriptions stay at the level of function and usage, without reporting how each example performs or where it applies; the two minimal toolset examples are positioned as interface demonstrations, and the native macOS implementation as a teaching reference.