v0.12.0More MCP/Docs
guides•Last updated: 2026-09-02

Architecture & Core Concepts

Understand how More MCP handles JSON-RPC 2.0, Streamable HTTP, the WordPress Abilities API, and local execution.

Architectural Philosophy

More MCP is designed around three strict principles:

  1. Native Execution: Code runs directly inside your WordPress PHP runtime, utilizing native WordPress functions (wp_insert_post, get_post_meta, etc.) rather than external database queries.
  2. Zero Middleware: Communication is direct between your AI client and your WordPress host. No cloud proxy, no relay servers, and zero vendor telemetry.
  3. Strict Boundaries: Every action is verified against WordPress capabilities, nonces, and destructive confirmation flags.

Protocol Layer: MCP 2025-11-25

More MCP implements the modern 2025-11-25 revision of the Model Context Protocol:

  • Transport: Streamable HTTP (Server-Sent Events / SSE for continuous streams and standard POST requests for stateless calls).
  • Format: Strict JSON-RPC 2.0.
  • Capabilities Negotiation: The server declares tools, prompts, resources, and notification capabilities during the initial initialize handshake.

WordPress Abilities API (WordPress 6.9+)

In upcoming WordPress 6.9 and modern environments, WordPress introduces the Abilities API, a formal core subsystem for registering AI-accessible tools.

More MCP bridges this ecosystem:

  • If WordPress 6.9+ Abilities API is present, More MCP registers its catalog with core automatically.
  • Any third-party plugin that registers core abilities has those abilities exposed to your AI client automatically.
  • On older WordPress releases (6.0 to 6.8), More MCP’s standalone registry provides the exact same tool definitions without requiring external dependencies.

Tool Classification: Read vs Destructive

Every tool in More MCP belongs to one of two categories:

  • Read-Only (81 baseline tools): Operations that query data, list items, inspect server health, or check options. These never modify state and can run without interactive user intervention.
  • Destructive (26 baseline tools): Operations that create, update, trash, delete, or rewrite content, options, users, or plugin files.

This strict division ensures that AI models do not accidentally overwrite production data without explicit confirmation.

Previous← Client SetupNextPermissions & Confirmation →