miniOrange

miniOrange Secure MCP Server helps WordPress administrators with AI governance and policy enforcement: understanding, and controlling, what AI assistants and MCP clients are allowed to do on their site.
The WordPress Abilities API (available in WordPress 6.9 and later) lets plugins and WordPress core expose discrete, machine-callable capabilities — for example: get site info, create a post, or generate a summary. This plugin turns those abilities into a remote Model Context Protocol (MCP) server so AI clients can discover and invoke them, protected by a self-hosted OAuth 2.1 authorization server.
What this version does
<ul>
<li>Abilities viewer. A read-only admin screen (in the Secure MCP Server menu) that lists every ability registered on your site, with its label, description, category, source namespace, and full input/output JSON schema.</li>
<li>Connection guide. A “Connect to AI” tab with step-by-step instructions and your site’s MCP URL for connecting clients such as ChatGPT and Claude.</li>
<li>Built-in content abilities. Create Post and Update Post abilities (exposed as MCP tools) so connected clients can draft and edit posts, gated by the user’s capabilities.</li>
<li>MCP server. A single Streamable HTTP endpoint that exposes every registered ability as an MCP tool. Tool calls run through the Abilities API, so each ability’s own permission check still applies.</li>
<li>Self-hosted dynamic OAuth. WordPress acts as its own OAuth 2.1 authorization server with OAuth 2.0 Dynamic Client Registration (RFC 7591), Protected Resource Metadata (RFC 9728), Authorization Server Metadata (RFC 8414), and Authorization Code flow with PKCE. Clients such as ChatGPT and Claude can register themselves and connect with no manual credential setup.</li>
</ul>
Every MCP request runs as the WordPress user who authorized it, so what an AI client can do is bounded by that user’s own capabilities.

miniOrange Secure MCP Server

AI governance for WordPress: expose your Abilities API as a secure, OAuth-protected MCP server for AI clients like ChatGPT and Claude.

Haktan Suren

HandL AI Connector Access Control lets administrators allow/deny AI Client prompt execution on a per-plugin basis using the <code>wp_ai_client_prevent_prompt</code> filter introduced with the WordPress AI Client.
Default behavior is allow.
Learn mode (Audit & log tab) logs every AI Client call without blocking, so you can discover callers before enabling deny rules on the Plugin rules tab.
Emergency kill switch blocks all AI Client calls except plugins you list as exceptions.
Caller attribution is best-effort and is determined by inspecting the PHP call stack and mapping file paths to installed plugins.
<h3>Privacy / Data</h3>
This plugin does not send data to any external service.
If you enable recent-call logging in Settings → HandL AI Connector Access Control, it stores a local log in the WordPress options table containing:
<ul>
<li>Timestamp</li>
<li>Allow/deny decision</li>
<li>AI Client operation (e.g. <code>generate_text</code>, <code>is_supported_for_text_generation</code>)</li>
<li>Provider and model when set on the prompt builder (or model preferences)</li>
<li>Truncated prompt preview and selected generation config (best-effort)</li>
<li>Input and output token counts when the AI Client completes a generation (best-effort)</li>
<li>Best-effort calling plugin (plugin basename) and source file</li>
<li>Current user id and display name</li>
<li>Request URI</li>
</ul>
Logs are kept as a count-based ring buffer (default 200 entries, configurable 20–1000). There is no time-based TTL—older rows drop only when the buffer is full.
Logging is disabled by default.

HandL AI Connector Access Control

Control which plugins may execute prompts via the WordPress AI Client.

miniOrange Secure MCP Server vs HandL AI Connector Access Control