Agentic MCP server for WordPress Elementor — multi-site management, safe Elementor data editing, template import/export, CSS flush, version tracking. Built for agencies running many client sites.
Server demonstrates solid foundational quality with 34 tools, all with descriptions and proper Zod schema validation. Naming follows verb_noun conventions consistently (list*, read*, update*, delete*, etc.). However, parameter descriptions are sparse or absent in many tools, the rubric requires descriptions for EVERY parameter, and this server falls short. Output schemas are partially defined (some tools have outputSchema, others do not). Error handling is basic (catches and logs errors, returns text responses) but lacks actionable recovery guidance. Tool annotations are present but underutilized, most tools lack readOnlyHint, destructiveHint, or idempotentHint flags despite clear risk classifications in the prompt. The server handles pagination implicitly but does not expose pagination parameters (limit, offset, page) in visible tool schemas, this risks returning unbounded result sets. Overall: solid naming, descriptions present but parameter-level docs weak, schemas half-done, error guidance absent, annotations incomplete.
Add a new widget to a container
Apply a template to an existing page
Find and replace text across all Elementor pages on a site
Check Elementor plugin versions across sites
Compare visual differences between page screenshots
Delete a widget from a page
Duplicate an Elementor page with all its data
Duplicate a widget on a page
Parameter descriptions missing or sparse across all 34 tools. Rubric requires non-empty descriptions for EVERY parameter to enable LLM understanding of what each param controls. Currently, many Zod schemas likely have parameters without descriptions, making it impossible for LLMs to determine valid values or constraints.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2025-06-18+ | v2 |
Export a page or template as an Elementor template file
Find and replace text in widget settings on a page
Find and replace text across Elementor pages on multiple sites
Import an Elementor template file
List available backups of Elementor page data
List all pages with Elementor data on a site
List all global widgets available on a site
List configured WordPress sites
List available Elementor templates
List all widgets in an Elementor page
Move a widget to a different parent container
Test connectivity to a WordPress site
Validate Elementor data integrity before editing
Read the Elementor data structure of a page
Read settings of a specific widget
Restore a page from an Elementor backup
Restore Elementor page data from a backup file
Take a screenshot of a rendered Elementor page
Get health status of a WordPress site
Replace a widget with a different type
Update settings of a specific widget
Execute a WP-CLI command on a site
Clear Elementor CSS cache
List installed WordPress plugins
Update a WordPress plugin
Search and replace in WordPress database via WP-CLI
Tool annotations incomplete. Risk classifications exist in prompt (READ_ONLY, WRITE, DESTRUCTIVE, IRREVERSIBLE) but are not surfaced in tool definitions. Tools lack readOnlyHint, destructiveHint, and idempotentHint flags. This prevents agents from reasoning about state mutation and replay safety. E.g., deleteWidgetTool, wpCliRunTool, wpSearchReplaceTool should declare destructiveHint: true or readOnlyHint: false.
Output schemas missing for many tools. The rubric requires documented return types for 100% of A+ tools. Currently, many tools do not declare outputSchema in their definitions. List tools (listSitesTool, listElementorPagesTool, listGlobalWidgetsTool) should document what structure they return (array of objects with which fields?).
Error handling lacks recovery guidance. Server.ts catches errors and returns plain text error messages (line: 'content: [{ type: "text" as const, text: `${tool.name} failed: ${(e as Error).message}` }]'). These do not tell agents what to do next, which step failed, or if the error is retryable. Errors should include recovery hints per pattern:recovery-guide.
Pagination parameters not visible in tool schemas. List tools (listSitesTool, listElementorPagesTool, listWidgetsInPageTool, etc.) should accept limit, offset/page, and return total_count or next_cursor. Without pagination controls, agents risk requesting unbounded result sets that exhaust context windows. Rubric requires limit 20-50 with documented constraints.
Destructive operations lack confirmation or dry-run support. wpCliRunTool (IRREVERSIBLE), deleteWidgetTool (DESTRUCTIVE), and wpSearchReplaceTool (database-wide WRITE) should support a dry_run or confirm_before_execute pattern to prevent catastrophic agent mistakes.
Tool naming could be more consistent. Some tools use gerund-based names (e.g., 'duplicateElementorPageTool') while the verb_noun pattern would be clearer ('duplicatePage'). While current naming is acceptable, tools like 'wpCliRunTool' and 'wpElementorFlushCssTool' are slightly vague, 'wpCliExecuteCommand' and 'wpFlushElementorCache' would be more actionable.
Dependency hints missing. Tools like 'readPageElementorTool' should hint that you must know the site_id and page_id first, maybe 'Start with listElementorPagesTool to discover available pages.' Similarly, 'updateWidgetSettingsTool' should note 'First call readWidgetTool to see current settings.'
No per-tool rate limiting or timeout configuration visible. wpCliRunTool, screenshotPageTool, and remote API calls could hang indefinitely. Server should implement explicit timeouts and rate limits to prevent runaway agents from overwhelming WordPress instances.