What if every website could expose its features as structured tools for AI agents - no backend MCP (Model Context Protocol) server required? WebMCP is an emerging W3C specification that lets web developers register JavaScript functions as discoverable, schema-validated tools directly in the browser. In this session, you'll go from zero to a working WebMCP integration: registering imperative and declarative tools, defining input schemas, handling agent-driven execution alongside normal user flows, and managing tool lifecycles. In a live demo, we'll walk through a tool-enabled page, inspect it with developer tooling, and trigger it from a browser agent. You'll leave with production-ready patterns and the open-source WebMCP AI agent skill I created that scaffolds correct integrations in seconds.

Learning Outcomes:
- Understand what WebMCP is, how it differs from server-side MCP, and why the browser is a natural tool-hosting surface for AI agents.
- Register, describe, and unregister imperative and declarative tools with proper schemas, annotations, and signal-based lifecycle management.
- Design tool descriptions and input schemas that guide agents toward correct invocation without brittle prompt engineering.
- Validate tool behavior deterministically with preview developer tooling before testing with a live agent.
- Bootstrap new WebMCP integrations faster using the open-source WebMCP AI agent skill and its detection scripts, templates, and reference materials.

Target Audience:
- Frontend and full-stack web developers who want to make their web applications agent-ready. No prior experience with MCP, AI agents, or browser AI APIs is required.

Suggested Keywords:
- WebMCP, Model Context Protocol, browser AI, AI agents, web standards, web AI, web platform

Outline:

- The problem: agents vs. the web today
-- How agents currently interact with websites: scraping, brittle selectors, screenshot parsing
-- Why structured tool discovery beats reverse-engineering the DOM
-- Where WebMCP fits: MCP-inspired, but in-browser and page-hosted

- WebMCP fundamentals
-- The exposure model: secure window context, no server required
-- Anatomy of a tool: name, description, input schema, execute callback, annotations
-- Imperative vs. declarative: when to use each shape
-- How agents discover, select, and invoke registered tools

- Live demo: from a plain webpage to an agent-operable app
-- Registering a first imperative tool on an existing web page
-- Defining a JSON Schema for structured input and validating inside the callback
-- Managing tool lifecycle: registering on mount, unregistering on teardown
-- Inspecting the registered tool set with the Model Context Tool Inspector
-- Triggering the tool from a browser agent and observing the round-trip

- Authoring tools that agents actually use well
-- Writing descriptions that reduce misrouting and hallucinated parameters
-- Schema design: explicit enums, raw user values, and loose-schema-strict-code validation
-- Returning results after the UI updates so agents can verify the effect
-- Common pitfalls: duplicate names, circular schemas, stale registrations

- Accelerating adoption with the WebMCP AI agent skill
-- What the open-source webmcp skill provides: detection scripts, registration template, reference docs
-- Running the workspace scanner to find integration targets
-- Adapting the template to your framework and shipping a first tool in minutes

- What's next and Q&A
-- Specification status: W3C Machine Learning Community Group draft, browsers preview, evolving declarative surface
-- Signals to watch: new browser support, declarative spec stabilization, accessibility considerations
-- Resources: spec link, skill repository, community group