What problem does it solve?
Browser AI agents currently interact with web apps by guessing at the DOM, which is fragile and error-prone. This Skill makes an existing web application agent-ready by exposing its user-facing actions as structured WebMCP tools (document.modelContext), with a human approval gate and real-browser verification so nothing ships unproven.
Core Features & Use Cases
- Phased pipeline with human gate: Runs DETECT, INVENTORY, INTEGRATE, VERIFY, HEAL, and AUDIT phases, pausing at a manifest approval gate where you approve or reject each proposed tool before any code changes.
- Declarative and imperative integration: Annotates standard HTML forms with toolname/tooldescription attributes, or registers imperative tools via a vendored, dependency-free runtime that wraps the app's existing UI code paths.
- Real-browser verification and self-healing: Generates Playwright specs that run in headed Chrome against the production WebMCP surface, then diagnoses and fixes failures with a bounded heal loop.
- Use Case: Point it at a React SaaS app and say "webmcpify the tickets area" — it maps user actions into a tool manifest, gets your approval, integrates the tools without touching unrelated code, and proves each one works in a real browser.
Quick Start
Run the webmcpify skill on this repository to inventory user-facing actions and expose them as WebMCP tools.
