MCP server for WordPress API integration with natural language processing capabilities
This server has severe quality gaps across naming, descriptions, schemas, and error handling. While it offers 36 tools covering WordPress CRUD operations, most lack proper documentation and structured parameter schemas. Tool names are reasonable (get*, create*, update*, delete* pattern), but input schemas are minimally defined with missing type constraints and incomplete descriptions. Parameter descriptions are sparse or generic. No output schemas are documented. Error handling lacks recovery guidance. The `processNaturalLanguageRequest` tool is particularly problematic, it accepts plaintext credentials as parameters (security violation) and has a vague description. The server is not built as a proper MCP server but as a REST API with an Express backend, raising fundamental protocol readiness concerns.
Create a new category in WordPress
Create a new comment in WordPress
Create a new page in WordPress
Create a new post in WordPress
Create a new tag in WordPress
Create a new user in WordPress
Delete a category from WordPress
Delete a comment from WordPress
Delete a media item from WordPress
Critical: Plaintext credentials exposed as tool parameters. `processNaturalLanguageRequest` accepts username, password, and app_password as direct parameters, violating the secret-injection pattern. These credentials will be logged in traces and potentially exposed in LLM context.
Critical: Missing input schemas for most tools. Tools like getPosts, getPages, getMedia, getUsers, getCategories, getTags, getComments, getMenus declare 'params' or data parameters but do not provide nested property definitions, type constraints, or validation rules. This prevents the LLM from understanding what values are allowed.
High: Generic or minimal parameter descriptions. Many tools have parameter descriptions like 'Optional query parameters' (getPosts, getPages) or just 'Tag data' (createTag), providing no guidance on what fields, formats, or constraints apply. LLMs cannot infer valid inputs from such vague guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 27 | - | v1 |
Delete a page from WordPress
Delete a post from WordPress
Delete a tag from WordPress
Delete a user from WordPress
Get all categories from WordPress
Get a specific category by ID from WordPress
Get a specific comment by ID from WordPress
Get all comments from WordPress
Get all media items from WordPress
Get a specific media item by ID from WordPress
Get all menus from WordPress using WP API Menus plugin
Get a specific page by ID from WordPress
Get all pages from WordPress
Get a specific post by ID from WordPress
Get all posts from WordPress
Get a specific tag by ID from WordPress
Get all tags from WordPress
Get a specific user by ID from WordPress
Get all users from WordPress
Process MCP-formatted requests with messages array or legacy query format
Process a natural language request and convert it to WordPress API operations
Update an existing category in WordPress
Update an existing comment in WordPress
Update an existing page in WordPress
Update an existing post in WordPress
Update an existing tag in WordPress
Update an existing user in WordPress
Upload a media file to WordPress
High: No output schemas documented. Tools return data but provide no description of response structure, field names, or types. LLMs cannot plan downstream tool calls or extract needed fields without knowing what the response contains.
High: Vague tool descriptions for data mutation tools. Tools like createTag, updateTag, createComment, updateComment have descriptions under 20 characters or lack clarity on what state changes occur, making it hard for LLMs to decide when to call them or understand consequences.
High: No error handling guidance. Source code shows error responses but no evidence of actionable error messages, recovery hints, or categorization (retryable vs. user-fixable vs. fatal). An LLM receiving a 500 error has no guidance on what to do next.
Medium: Credentials not abstracted from tools. While the server reads WordPress credentials from environment variables (WP_URL, WP_USERNAME, WP_APP_PASSWORD in config.js), the `processNaturalLanguageRequest` tool allows agents to override these with user-supplied credentials, re-opening the credential exposure risk.
Medium: No pagination guidance for list operations. Tools like getPosts, getPages, getUsers that retrieve 'all' items lack descriptions of pagination behavior, limits, or offset/limit parameters. Large result sets can blow context windows without pagination.
Medium: Destructive tools lack confirmation/dry-run support. deletePost, deleteUser, deleteCategory, deleteTag, deleteComment, deleteComment, and deleteMedia can permanently remove data but provide no confirmation, dry-run, or undo mechanism to prevent agent mistakes.
Low: Tool naming for processNaturalLanguageRequest is unclear. The verb 'process' is generic; a more specific name like 'execute_wordpress_operation_from_text' or 'interpret_and_execute_query' would better convey intent.