🌐 WebMCP: How We Made Ghost Blog Directly Accessible to AI Agents

The Problem: AI Agents Shouldn’t Need Browsers

Today, when an AI agent needs information from a website, the typical flow looks like this:

1. Launch a headless browser
2. Navigate to the page
3. Parse the DOM / read rendered UI
4. Extract the data it actually needs

This is slow, brittle, and expensive. The agent is doing what a human would do — but it’s not a human.

The Inspiration

We were inspired by skirano’s WebMCP demo, which showed how “the browser becomes an API.” The idea clicked immediately:

What if our static blog could speak directly to AI agents — no browser required?

What We Built

We added four JSON endpoints to our Jekyll (GitHub Pages) site, turning it into a WebMCP-compatible data source:

1. /.well-known/mcp.json — Discovery

{
  "name": "Ghost Blog",
  "description": "AI and automation blog",
  "endpoints": [
    "/api/articles.json",
    "/api/articles-full.json",
    "/api/meta.json"
  ]
}

This is the entry point. An AI agent hits /.well-known/mcp.json and instantly knows what APIs are available — just like robots.txt tells crawlers what to index.

2. /api/articles.json — Article Listing

Returns a lightweight list of all posts with metadata:

• title, url, date, description, author, tags

Perfect for agents that need to browse or search content without downloading full articles.

3. /api/articles-full.json — Full Content

Same as above, but includes the complete article body in each entry. For agents that need to ingest, summarize, or reference the actual content.

4. /api/meta.json — Site Metadata

Site name, authors, total article count, and other structural info. Useful for agents building knowledge graphs or doing site-level analysis.

How It Works (Technically)

Here’s the key insight: Jekyll can generate JSON files using Liquid templates.

We don’t have a server. We don’t have a database. We don’t have an API framework. We have _posts/ and Liquid.

Each JSON endpoint is a file like api/articles.json with YAML front matter and Liquid logic:

---
layout: null
---
[
{% for post in site.posts %}
  {
    "title": {{ post.title | jsonify }},
    "url": "{{ site.url }}{{ post.url }}",
    "date": "{{ post.date | date_to_xmlschema }}",
    "description": {{ post.description | jsonify }},
    "author": {{ post.author | jsonify }},
    "tags": {{ post.tags | jsonify }}
  }{% unless forloop.last %},{% endunless %}
{% endfor %}
]

Every time we push a new post, GitHub Pages rebuilds the site and the API updates automatically. Zero maintenance. Zero cost.

Before vs. After

Why Now?

The web’s audience is changing. Increasingly, the “readers” of web content are not humans scrolling through pages — they’re AI agents fetching, indexing, and reasoning over data.

If your content isn’t machine-readable, it’s invisible to the next generation of the web.

By publishing structured, AI-friendly endpoints now, we’re positioning this blog as:

• A knowledge base that agents can directly reference
• An early mover in the WebMCP ecosystem
• A zero-cost example that any Jekyll/GitHub Pages site can replicate

Try It Yourself

If you run a Jekyll site, you can implement WebMCP in under an hour:

1. Create /.well-known/mcp.json with your endpoint list
2. Add Liquid-templated JSON files under /api/
3. Push to GitHub Pages
4. Your blog now speaks to AI agents

The future web isn’t just for humans. Make sure your content is ready.

This blog itself is WebMCP-enabled. AI agents can discover our endpoints at /.well-known/mcp.json and access all articles programmatically.