WordPress Plugins
Free Tools
Claude Skills
Pricing Blog Case Studies Switch to Royal Plugin Graveyard Support My Account Cart
Support / Royal MCP / Using with Claude Desktop

Using Royal MCP with Claude Desktop

You’ve connected Claude Desktop to your WordPress site through Royal MCP — now what? This guide covers the ten starter prompts that make an agent immediately useful, the read-before-write habit that separates good sessions from broken ones, common multi-step workflows for content, WooCommerce, SEO and troubleshooting, and how to read the Activity Log when something surprises you.

✓ This guide picks up after setup is done

Not connected yet? Start with the Connecting to Claude walkthrough — it covers claude.ai web, Claude Desktop, and Claude Code CLI in one place. Once you can ask Claude “what plugins are on my site?” and get a real answer, come back here.

Ten Starter Prompts

These are the first prompts we suggest new customers try. They’re grouped by intent so you can see which tools an agent will reach for — and each one is copy-pasteable. Replace example.com with your actual site domain.

Site inspection (start every session here)

  1. “On example.com, run wp_get_site_status and give me a one-line environment summary.” — single call collapses WP + PHP + MySQL versions, active plugin count, memory limit, disk free, install age. Best opening move on a debugging call.
  2. “On example.com, list my installed plugins and flag any that look out of date or abandoned.” — Claude cross-references version numbers against what it knows about WP plugins.
  3. “On example.com, tail the last 50 lines of the WP error log and summarize what’s recurring.” — needs WP_DEBUG_LOG enabled; Claude will tell you if it isn’t.

Content workflows

  1. “On example.com, show me my five most-recent draft posts and pick the one closest to shippable.” — agent reads, ranks, recommends. No writes.
  2. “On example.com, take draft post ID 412 and rewrite it in a tighter, more direct voice. Save as a new revision, don’t publish.” — the “save as revision, don’t publish” framing is important; it gives you a rollback point.
  3. “On example.com, find any published posts that mention ‘[old brand name]’ and generate a search-and-replace preview before touching anything.” — forces a read + report before the write.

WooCommerce (if active)

  1. “On example.com, list all coupons that expire in the next 30 days.” — Claude filters via wc_get_coupons without needing a plugin dashboard.
  2. “On example.com, show me every product with fewer than 5 units in stock, grouped by category.” — two reads + local grouping; no writes.

SEO & term meta

  1. “On example.com, list categories that have no SEO description set (Yoast, Rank Math, or AIOSEO) and draft a short one for each.” — Claude auto-detects the active SEO plugin and reads the right term-meta key.
  2. “On example.com, show me my scheduled wp_cron events and highlight anything overdue.” — catches stuck cron before it becomes an incident.
Why every prompt starts with “On example.com”

If you’ve connected more than one WordPress site (a common Royal MCP pattern), Claude has to pick one on your behalf when a prompt is ambiguous. Naming the domain up front removes that ambiguity and cuts down on “which site did that go to?” moments later. This is the single highest-ROI habit for multi-connector users.

The Read-Before-Write Habit

The biggest quality gap between beginner and advanced Royal MCP sessions is a single habit: reading before writing. Agents that jump straight to a write call — overwriting a post, updating a coupon, deleting a term — produce more surprises than agents that first read the current state, report it back to you, and only then propose a change.

You don’t have to enforce this manually. Two lines added to any editing prompt will do it:

The universal editing prompt suffix

“Before you write anything, read the current state and show me what you’re about to change. Wait for me to confirm.”

Concrete example. Instead of:

Update the excerpt on post 412 to "New excerpt copy here."

Prefer:

On example.com, read post 412's current title and excerpt.
Show me what's there today and what you'd change it to.
Wait for me to confirm before writing.

The extra 15 seconds catches wrong-post-ID mistakes, prevents overwriting a working excerpt with an inferior one, and gives you the confirm/reject checkpoint that AI-agent editing is missing by default. It also matches how the underlying MCP tools were designed — Royal MCP’s read tools are cheap and fast; use them.

When the pattern matters most

Common Multi-Step Workflows

These are patterns that turn into single Claude prompts but involve three-to-six chained tool calls under the hood. All of them work today — try them in your own site’s conversation.

“Publish this draft with everything set up right”

Prompt: “On example.com, take draft post 412, assign it to the ‘Announcements’ category, set the featured image to the image with alt text ‘Team photo Q3’, and publish. Show me each step before you take it.”

Under the hood: wp_get_postwp_get_categories (find ID) → wp_get_media (find alt-text match) → wp_set_featured_imagewp_add_post_termswp_update_post (status=publish). Six calls, one prompt.

“Rebuild this WooCommerce coupon set for a Q4 sale”

Prompt: “On example.com, list all coupons with codes starting ‘FALL2025’, delete them, then create fresh ‘FALL2026’ equivalents with the same discount rates, expiring 2026-11-30. Preview the delete + create plan first.”

Under the hood: wc_get_coupons → (agent shows preview + waits) → wc_delete_coupon × N → wc_create_coupon × N.

“Audit my categories for SEO description coverage”

Prompt: “On example.com, walk through every category, check the Rank Math description meta, and give me a table of categories missing descriptions with a suggested one-sentence description for each.”

Under the hood: wp_get_categories → per-category wp_get_term_meta → local scoring → markdown table output. No writes unless you follow up.

“Something’s broken — run a full diagnostic”

Prompt: “On example.com, run a full health check: site status, error-log tail (last 100 lines, filter ‘Fatal’), cron schedule with anything overdue highlighted, and installed plugin list. Summarize what looks wrong.”

Under the hood: wp_get_site_status + wp_get_error_log_tail + wp_get_cron_schedule + wp_get_plugins, all in parallel. This is the diagnostic-first-response pattern — run it before you start guessing.

When Something Surprises You — Read the Activity Log

Royal MCP logs every tool call in WP Admin → Royal MCP → Activity Log. When an agent’s answer doesn’t match what you expected, this is where you look first.

Each row shows: timestamp, tool name, arguments, return value, and success/error status. A few patterns to know:

Claude says a change went through — did it really?

You asked Claude to update a post excerpt. It reported success. You look at the post and the excerpt is unchanged.

Check the log

Find the wp_update_post row (or wp_update_post_meta depending on what field). Look at the arguments — is the post ID right? Look at the return — does result equal true? If arguments and return both look correct, the write happened; your browser is caching the old view (hard-refresh with Ctrl+Shift+R).

“Session credentials mismatch” 403 after an update

Everything was working, you updated Royal MCP, and now Claude gets 403 errors on the first tool call.

Cause + fix

Your MCP session was bound to the previous plugin version’s session-fingerprint logic. Restart your Claude Desktop app (or open a new claude.ai chat / run /mcp → Authenticate in Claude Code) — the fresh session re-initializes cleanly against the current version. Pre-1.4.35 sessions self-heal within 24 hours; post-1.4.35 sessions survive plugin updates automatically.

Claude ran a tool you didn’t expect

You asked for a list of posts; Claude also ran wp_get_categories and wp_get_users before answering.

Why

Agents will sometimes gather adjacent context to give a better answer. That’s desired behavior on ambiguous prompts. If you want strictly minimal calls, ask: “Answer with a single tool call. No exploratory reads.” Claude will comply.

What’s Next

The above covers the day-to-day Royal MCP + Claude Desktop workflow. If you want to go deeper: