Why Signett

Web applications already contain precise capabilities: search inventory, retrieve an invoice, schedule a meeting, change a setting, or cancel an order. Agents usually reach those capabilities indirectly by reading a human interface and clicking through it.

WebMCP gives cooperating applications a better primitive: register a structured tool in the page and let a compatible agent discover and invoke it. Signett is the developer workflow around turning application capabilities into useful WebMCP tools.

The missing workflow

A raw registration call is intentionally small. Shipping a useful agent interface still requires product and engineering decisions:

• choose reusable capability boundaries;
• write names and descriptions agents select correctly;
• validate untrusted arguments at runtime;
• expose each tool only in the right page and application state;
• inspect the exact schemas and annotations an agent sees;
• test discovery, lifecycle, invocation, errors, and outcomes;
• add stronger controls when a tool can cause a consequential effect.

Signett aims to make that loop coherent without replacing WebMCP or taking ownership of application logic.

browser agent
    │
    ▼
native WebMCP tool
    │
    ▼
Signett definition, exposure, and test workflow
    │
    ▼
existing application function and backend

Progressive hardening

A public product lookup should stay simple. An authenticated invoice read may need application context and authorization. A state-changing action may also need durable idempotency and authoritative outcome verification.

The createSignett().expose() workflow adds those controls to individual tools as needed. The lower-level guard() remains available for an existing native registration, but most applications should start with the interface API.

What Signett is not

Signett is not a new protocol, a production browser polyfill, a remote tool catalogue, or a hosted replacement for your application. Native WebMCP remains visible, and your identity, policy, data, handlers, and backend stay authoritative.

Next: choose an outcome and define its proof with the User Jobs to Be Done workflow.