Human-authorized execution for WebMCP

Let the agent move. Keep the decision yours.

doA2Ai is a browser-native collaboration pattern for the moment when useful preparation becomes a consequential action.

The design question

How can an agent be genuinely helpful without quietly becoming authorized?

Scope before speed

The agent starts with only the context and capabilities that belong to the work in front of it—nothing quietly follows it from task to task.

Permission at the point of consequence

Preparation can stay fluid. When an action changes the world, the person sees the exact proposed move and decides there.

A receipt, not a shrug

Afterward, the system can show what was requested, what was approved, and whether the outcome still matched the boundary.

A simple handoff

Useful automation, with a visible seam.

Context
Work inside the present task.
A browser surface exposes the relevant tools and information for this one piece of work.
Decision
Make the consequential choice explicit.
A person is invited in only when the scope changes or an exact approval is needed.
Receipt
Leave something inspectable behind.
The outcome can be checked against the permission that made it possible.

Why it matters

The important interaction is not a chatbot prompt. It is the boundary between preparation and permission.

doA2Ai is a WebMCP prototype exploring a more legible relationship between people and browser-based agents: no invisible standing authority, no performative approval queue, and no ambiguity about who made the call.

The agent can help with the work. The person remains the source of authority.