WebMCP: Turn Your Website Into a Tool AI Agents Can Actually Use

Right now, an AI agent works your website the way a human does badly. It reads the pixels, guesses where a button sits, clicks, then checks what happened and repeats. WebMCP, a new standard from Microsoft and Google inspired by Anthropic's Model Context Protocol, gives agents a direct way to use your site's tools instead of clicking around it.

Kowalah has already shipped three WebMCP tools on its live AI Programme Cost Calculator. In this session Charlie shows an AI agent setting parameters, pulling a cost estimate, and asking for the maths behind it, talking straight to the page instead of screenshotting it.

What you'll learn:

• The difference between MCP (agent to your applications) and WebMCP (agent to your website)
• Why the 'front door' (WebMCP) and the 'back door' (a remote MCP server) solve different problems
• When a live, logged-in human session beats an API key and an OAuth flow
• The exact three tools Kowalah built for its cost calculator, and what each one does

Who should attend:

Anyone whose website makes a human do something: book, apply, self-diagnose, or fill in a multi-step form. If your site is a brochure with a contact form, this will interest you less. If it runs a process, it's worth 45 minutes.

What you'll leave with:

• A working definition of WebMCP you can explain to your own team
• A checklist of questions to run against your own website's flows
• A live example to point colleagues to when you propose testing this