— Native MCP Server —
Your WordPress site, as a first-class AI data source.
Most “AI-ready” plugins drop a static llms.txt file at your root and call it done. Schema Monkee turns your site into a live Model Context Protocol (MCP) provider — the same interface Claude Desktop, Cursor, and every modern agent platform already speak. When an AI needs to know something about your business, it queries you directly, in real time, with full entity resolution.
Primer
What’s the Model Context Protocol, and why should your WordPress site speak it?
MCP is an open standard — introduced by Anthropic in late 2024, now adopted across the major agent platforms — for letting AI agents query live data sources the same way they query their own tools. An MCP server advertises resources (data you can read) and tools (actions you can invoke). A compliant client (Claude Desktop, Cursor, any MCP-aware agent) can then access them structurally, without scraping HTML or hoping the model’s training data is current.
Open standard
MCP is vendor-neutral — any agent platform can implement a client. Your site doesn’t get locked to one AI provider.
Structured, not scraped
Agents query your @graph directly — not your HTML. Entity resolution happens by @id, not by regex.
Live, not trained
The agent sees the version you just published — not a snapshot from the model’s training cutoff 18 months ago.
What We Publish
Two endpoints. Both auto-generated from your verified graph.
No hand-authoring. No drift. Both files regenerate on every Pass 3 webhook, so what the agents see is always what you just published.
GET /llms.txt
Human-legible summary for LLM crawlers.
A structured Markdown file summarizing your business, services, key people, and canonical @ids. Pre-parsed for LLMs that don’t speak MCP yet but do look at llms.txt when training or indexing. Generated from Pass 3 output, updated on every webhook.
# Acme Coffee Roasters
Specialty coffee roaster & cafe in Brooklyn, NY.
Founded 2019.
## Canonical Entities
- Organization: https://acme.coffee/#org
- Founder: https://acme.coffee/#founder
- Location: https://acme.coffee/#location
## Services
- Wholesale coffee supply
- Direct trade sourcing
- Retail cafe — 412 Bedford Avenue, Brooklyn NY 11249
## Authority
- Wikidata: https://www.wikidata.org/wiki/Q123456789
- Award: Roast Magazine Macro Roaster of the Year, 2024
## Contact
tel:+1-718-388-2000
email:hello@acme.coffeeGET /.well-known/mcp.json
Machine descriptor. This is the one agents query.
A strict-schema JSON document advertising the MCP resources your site exposes. Agents fetch this first to discover what they can ask you — your full @graph, your locations, your services, your team — then issue follow-up queries for the specific resource they need.
{
"mcpVersion": "2024-11-05",
"serverInfo": {
"name": "Acme Coffee Roasters",
"version": "2026.04.17",
"description": "Official MCP provider for acme.coffee"
},
"capabilities": {
"resources": { "list": true, "read": true }
},
"resources": [
{
"uri": "schema-monkee://acme.coffee/@graph",
"name": "Entity Graph",
"mimeType": "application/ld+json"
},
{
"uri": "schema-monkee://acme.coffee/locations",
"name": "Retail Locations",
"mimeType": "application/json"
},
{
"uri": "schema-monkee://acme.coffee/offers",
"name": "Products & Services",
"mimeType": "application/json"
}
]
}In Practice
From agent config to answered question, in two files.
Here’s the whole flow, end to end. An operator wires your site into Claude Desktop; a user asks a question; Claude resolves the answer by querying your graph directly.
01 Client config — Claude Desktop
// ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"acme-coffee": {
"url": "https://acme.coffee/.well-known/mcp.json"
}
}
}02 User asks Claude a question
User: "What awards has Acme Coffee Roasters won,
and who's the head roaster?"
[Claude invokes MCP tool: acme-coffee]
GET schema-monkee://acme.coffee/@graph
→ 200 OK, 2 entities returned
Claude: Acme Coffee Roasters was named
Roast Magazine's Macro Roaster of the Year, 2024.
Their head roaster is Jane Acme,
a Q Arabica Grader certified by CQI.
(Source: acme.coffee verified entity graph)Notice: Claude cites your graph as source — not “the web” or “my training data.” That’s the difference between being scraped and being queried.
What “Native” Means
The difference between shipping a file and running a service.
The “Static File” Approach
Hand-author once. Watch it rot.
- Authored manually — wrong the moment anything about your business changes
- No live query — agents download a static snapshot
- No entity resolution —
@ids either missing or not honored - No resource catalog — agents can’t discover what else you expose
- No sync with your
@graph— the two drift apart immediately
Schema Monkee Native MCP
Auto-generated. Queryable. Always current.
- Generated from Pass 3 output — every webhook refreshes it
- Live MCP service — agents issue real-time queries
- Full entity resolution —
@ids resolve, relationships traverse - Resource catalog advertises everything exposed — graph, locations, offers
- Single source of truth — your MCP surface is your graph
Already have an llms.txt? We’ll show you what it’s missing.
A free audit checks whether your site already advertises an MCP surface, whether any agent can resolve your canonical entities, and what a native MCP provider would expose that yours doesn’t today.