atifsheraz

Hi @jayson35

First, thank you again for the quick ChatGPT compatibility fix. I installed the 1.5.3 hotfix and can confirm that ChatGPT is now successfully authenticating and using Royal MCP on my WordPress site.

After using the integration more extensively, I wanted to suggest a possible direction for Royal MCP.

Royal MCP already has an excellent WordPress tool surface. On my current site ChatGPT can see more than 200 Royal MCP tools, and the self-hosted architecture, OAuth support, capability checks, logging, plugin integrations and undo functionality are major strengths.

The area where I think Royal MCP could become substantially more powerful is the development workflow around those tools.

Products such as WPVibe are increasingly providing an AI workflow where an agent can work on a site in an isolated draft, inspect/edit theme files, validate the result, generate a private frontend preview, show the human what changed, and only then promote the changes.

I would love to see Royal take this concept in its own security-first, self-hosted direction rather than simply allowing unrestricted filesystem or shell access.

The five capabilities that would make the biggest difference to me are:

1. Sandboxed theme development workspace

Create an isolated copy/workspace of the active theme or child theme where an AI can safely list, search, read and patch files without modifying the live theme.

Ideally this would include PHP syntax validation, strict path restrictions, file-size/type restrictions, before/after hashes and a changed-file manifest.

The live theme would remain untouched until an explicit publish/promote operation.

2. Diff + private preview + explicit publish

After modifying the workspace, let the AI return a structured diff and generate a private frontend preview URL.

A human could inspect the actual site and explicitly approve publishing the workspace.

Royal’s existing undo functionality could then protect the post-publish side as well.

This would give a very strong workflow:

draft → validate → diff → preview → approve → publish → undo if necessary.

3. Compact/layered tool discovery

Royal’s tool coverage is becoming very large. My current ChatGPT connection exposes over 200 tools.

It may be worth offering an optional compact MCP profile inspired by WordPress’s own MCP Adapter architecture, where the client initially receives a small number of tools for discovering capabilities, inspecting a tool schema and invoking it.

For example:

• discover_tools
• get_tool_info
• execute_tool

with filtering by category/plugin/read-only/write/destructive.

The existing direct-tool endpoint could remain available.

This may reduce context usage and improve tool-selection reliability as Royal continues adding integrations.

4. MCP Apps approval/change-review interface

Since MCP Apps is now an official MCP extension and is supported by clients including ChatGPT and Claude, Royal could potentially render a change-review panel directly inside the conversation.

For example:

• files/options/content being changed
• before/after diff
• validation results
• preview link
• Approve & Publish
• Discard

That would fit Royal’s security positioning very well.

5. General frontend verification + reusable skills

A generic rendered-page inspection tool would let the agent verify what WordPress actually produced after a change rather than only confirming that a database write succeeded.

It would also be useful if Royal’s existing Skills Index evolved into MCP-accessible task/site skills, so an agent could load instructions for workflows such as Gutenberg, GeneratePress, Elementor, WooCommerce, SEO, etc.

Site owners could eventually define their own persistent rules such as:

• always work in a draft first
• never publish without approval
• preserve the site’s design system
• validate frontend output after changes
• never modify the parent theme

Longer term, an opt-in safe WP-CLI-style administration layer and structured Gutenberg block-editing tools would also be very useful, but I think the draft workspace + preview/approval workflow would provide the biggest immediate improvement.

I especially like Royal MCP’s self-hosted model, so I would not want these features to weaken that architecture. The ideal result in my view would be the development convenience of an AI WordPress workspace while keeping Royal’s self-hosted, capability-gated and auditable security model.

Is anything along these lines already planned for Royal MCP / Royal MCP Pro? I would be happy to test an early implementation against a real WordPress development workflow if useful.

I’m also interested in long-term opportunities around projects like this, and I’d be happy to contribute where my testing, product feedback, WordPress experience, or implementation work could be useful to the project.