Websites have spent decades speaking to two audiences: humans and search engines.
There’s a third one now — AI agents.
When someone asks ChatGPT, Claude, or Perplexity to find a developer, compare services, or shortlist suppliers, an agent often visits your site first. If your website can’t communicate clearly with that agent, you risk being misrepresented or skipped.
Most sites still rely on the agent scraping HTML and guessing context. That works, but it’s messy.
I wanted something cleaner.
The problem with scraping-only discovery
Scraping gives you fragments: headings, snippets, and partial structure. It does not reliably provide intent.
For a business site, that’s a problem:
- Services can be interpreted incorrectly
- Pricing/context can be lost
- Contact and booking journeys can break
- Recommendations become vague
If an agent can’t confidently understand what you do, it’s less likely to recommend you.
What I implemented: WebMCP
WebMCP (Web Model Context Protocol) lets a website expose discoverable tools to AI agents.
In plain English: instead of forcing agents to infer from raw page content, you explicitly tell them what actions and queries are available.
I registered tools on my site so agents can:
- Search services
- Pull service details
- Browse portfolio projects
- List and read blog posts
- Send contact messages
- Request consultations
That means an agent can go from “find me a developer in Somerset for an e-commerce build” to a structured, accurate recommendation without guesswork.
What changed after implementation
The biggest difference is consistency.
Point an agent at a scraping-only site and you often get a rough summary. Point it at a site with structured tool discovery and responses become far more accurate.
This isn’t theoretical anymore — AI-driven discovery is already here. Search is no longer just “ten blue links”.
Lessons learned
A few practical things that mattered:
- Keep tools focused. One clear action per tool beats a catch-all endpoint.
- Return structured data. JSON objects outperform free-form blobs.
- Mark side effects correctly. Read-only vs action tools need clear boundaries.
- Test with real agents. Simulated tests miss edge cases.
Check your own site
I built a free scanner so you can check where your website stands:
It gives you a quick score and shows what AI agents are likely seeing right now.
Why this matters now
Google has dominated web discovery for years. AI agents are becoming an additional discovery layer on top.
Businesses that adapt early will be easier to find, easier to understand, and easier to recommend.
If your site still depends entirely on scraping, you’re competing with one hand tied behind your back.
If you want help making your website AI-agent ready, get in touch.