Skip to content

Tutorials

The tutorials move from one browser-local function to a production-shaped operation. Each one ends with observable evidence: a registered native tool, an execution trace, and—when the tool changes data—an authoritative application result.

Starting from your own application? Use the User Jobs to Be Done workflow to choose one outcome, time-box its capability sketch, define the Case and oracle, and only then implement the minimum tool surface. The codelabs below provide runnable examples for each proof lane.

CodelabWhat you build or runWhat it teachesTime
1. First agent callA React page with get_greetingRegistration, Chrome discovery, manual invocation, and your first signett agent run15 minutes
2. Headless agent testingA terminal-driven WebMCP regression suiteAd-hoc prompts, saved tasks, fresh browser Trials, authoritative grading, redaction, and automation25 minutes
3. Authenticated paymentThe repository's React and Express payment appThe complete loop: controls, browser and agent Trials, database grading, baselines, iteration, and change checks45–60 minutes
4. Cal.diy bookingA real booking through a local Cal.diy checkoutDiscovery-before-mutation, human confirmation, ambiguous-outcome recovery, and a Postgres oracle45–90 minutes

Start with the first codelab even if you already know React. It establishes the three systems involved in every example:

text
website + Signett -> native WebMCP tool -> agent runtime -> model

Signett defines and guards the website capability. The browser exposes that capability. An agent runtime connects a model to the browser. A model does not call a webpage by itself.

After the codelabs, use patterns from Cal.diy and Saleor as the integration checklist for your own application.

Your product capabilities, directly usable by agents.