The web is shifting from a place AI reads to a place AI acts. Your website is starting to not just be a brochure anymore. Rather, it's becoming a tool that AI agents use to get their patients booked correctly with you.
What actually happened
Google and Microsoft engineers built something called WebMCP — Web Model Context Protocol. It shipped in Chrome 146 Canary behind a feature flag. The idea is pretty simple once you cut through the jargon:
Right now, AI agents (think ChatGPT browsing, Gemini, Perplexity, browser agents, etc.) interact with websites by basically taking screenshots and guessing where to click or by scraping the raw HTML to try to understand the site.
WebMCP flips that. Instead of the AI guessing how to use your website, your website tells the AI exactly what tools are available. A practice website could expose a structured "book appointment" action that an agent can call directly, instead of the AI trying to figure out which blue button is the submit button.
Instead of an AI agent clicking around a practice website trying to find the appointment form, the site just says "hey, here's my booking tool, here are the inputs I need, done."
Why this matters for healthcare marketing websites
Back in December I wrote our "AI-friendly healthcare website strategy" doc. The whole premise was that making a website AI-friendly isn't about some secret robot translator code, it was simply doing the right things right.
To me, everything in that doc still holds. Structure your data. Connect your entities (doctor/location/services/insurance). Answer questions directly. Be the clear, trustworthy answer.
WebMCP just adds another layer to it.
What should you do right now?
Nothing…yet. This is early. No formal launch date. Industry insiders think we'll see formal announcements by mid-to-late 2026 at Google I/O or Cloud Next.
What can we do about it today?
Build a site that has:
- Clean, structured HTML (not bloated off-the-shelf WordPress themes)
- Schema markup and entity relationships baked in
- A CMS that is flexible enough to add these capabilities when the time comes
Synthesis
Google just sent a clear message that the future web is a place where AI agents not only read sites but use them. The lines get a little weird for healthcare and HIPAA reasons and I'm not sure where it will land. This will hit e-commerce first for sure. But the future state doesn't seem too far off.
Frequently Asked Questions
- What is WebMCP?
- WebMCP — Web Model Context Protocol — is an emerging standard for letting AI agents interact with websites in a structured way. Where llms.txt gives an LLM a static summary of a site, WebMCP gives it a way to call defined functions on the site (search, book appointment, find provider) through a structured contract. It is to AI agents what an API is to a developer.
- Does my healthcare website need to support WebMCP?
- Not yet, but soon. WebMCP is at the early-adopter stage — Google, Anthropic, and others are signaling support; production deployment is spotty. For multi-location healthcare groups, the relevant question is what AI agents will be able to do on behalf of patients in 12–18 months: book appointments, check insurance acceptance, find providers. Practices that expose these capabilities cleanly via WebMCP will be reachable by those agents; practices that don't, won't.
- How is WebMCP different from a regular website API?
- Two ways. First, the contract is designed for AI agents to consume and is human-readable in a way most APIs are not. Second, it is opinionated about discoverability — WebMCP-enabled sites publish their capability map at a known location, like robots.txt or llms.txt, so any AI agent can find and parse it without prior integration. APIs work for one-off integrations; WebMCP works for being agent-discoverable at scale.
- When should a healthcare group plan to add WebMCP support?
- Now if you are doing a major rebuild anyway. Otherwise, watch the space for the next 6–12 months. The standard is moving fast. Carenetic's guidance for multi-location groups: prioritize the foundations (clean schema, llms.txt, fast site) first; layer WebMCP in once the standard stabilizes and your portfolio's underlying booking/provider directory APIs are clean enough to expose.
More on AI & Search
- AI How to Make ChatGPT Love Your Website: A 5-Step Guide for Non-Techies
- AI How to Make ChatGPT Love Your Website: A 5-Step Guide for Non-Techies
- AI Modern Healthcare Search: Comparing ChatGPT Search, Google Search, and Perplexity AI
- AI LLMS.txt: The New Robots.txt for AI – What Healthcare Practices Need to Know
- SEO Zero-Click Searches and Your Physician Practice