Connecting WordPress to AI agents using Model Context Protocol
Executive Summary
Autonomous LLM agents (Claude Desktop, Cursor, LangChain) need standardized, bi-directional access to content management systems. We designed and open-sourced an implementation of the Model Context Protocol (MCP) for WordPress over JSON-RPC 2.0 with Server-Sent Events (SSE), introducing cryptographic SHA-256 capability scoping and dynamic tool discovery.
1. Why WordPress Needs a Model Context Protocol
WordPress powers more than 40% of the world’s websites. However, integrating modern autonomous AI agents with WordPress has historically required bespoke REST endpoints or brittle database scrapers. Standard REST APIs lack dynamic schema self-discovery and stateful execution streaming.
The Model Context Protocol standardizes how client agents discover available tools, request resource schemas, and execute commands safely with granular capability controls.
2. Protocol Architecture & Transport
Dynamic Tool Discovery (/mcp/v1/tools)
Returns complete JSON Schema definitions of WordPress capabilities (post authoring, media querying, SEO metadata tuning) without manual client prompt configuration.
Server-Sent Events (SSE) Streaming
Provides real-time feedback for long-running agent tasks (bulk image tagging, database optimization, content translations) without HTTP timeout drops.
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "wp_create_post",
"arguments": {
"title": "Autonomous Agent Architecture",
"status": "draft",
"categories": ["AI", "Engineering"]
}
},
"id": 1042
}Cite This Work & Download Specification
Download the complete RFC specification or explore the open-source GitHub codebase:
@techreport{kites2026wpmcp,
title={Decentralized Model Context Protocol: Cryptographic Tool Transport for Enterprise CMS},
author={Kites Protocol Working Group},
institution={Kites.Dev},
number={RFC-2026-04},
year={2026},
url={https://kites.dev/research/wpmcp-model-context-protocol}
}