WordPress MCP has 5 tools with moderate definition quality. Tool naming follows verb_noun convention (search-posts, get-post, create-post, update-post) which is good for LLM clarity. However, parameter schemas are minimal, most parameters lack proper type constraints and validation ranges. Descriptions exist but are uneven: search-posts and get-post are concise (good); create-post and update-post have excessively verbose descriptions (1400+ chars with embedded block format documentation that wastes tokens and confuses intent). The block-types-schema tool lacks a paramsSchema definition entirely (hardcoded empty object {}, violating schema registration patterns). Output schemas are not documented, LLMs cannot plan downstream calls because the response structure is opaque. Error handling is absent; tools will fail with bare exceptions instead of actionable recovery guidance. No input validation or mutually-exclusive parameter documentation. Overall, this reads as a basic integration with minimal LLM-optimization, it works but invites misuse.
Create a new WordPress post using Gutenberg blocks
Get a WordPress post by its ID
Search for WordPress posts by title or keyword
Update a WordPress post using Gutenberg blocks
Available WordPress block types that can be used for content creation
block-types-schema tool lacks input schema definition (paramsSchema = undefined). Tool registered without paramsSchema parameter to .tool() method (line 52 in index.ts). This violates schema registration patterns and forces LLMs to guess whether parameters are needed.
create-post and update-post descriptions are 1400+ characters, embedding verbose block format examples and attribute documentation inline. This wastes tokens, dilutes intent clarity, and violates token economy. Descriptions should be 50-200 chars; examples belong in a separate resource or interactive block picker tool, not in the parameter description string.
No output schemas documented. Tools return CallToolResult with unstructured text content. LLMs cannot parse response structure, plan chaining (e.g., get-post returns a post_id which likely downstream tools need), or extract field values reliably. Every tool should document the structure of its response.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 32 | - | v1 |
No error handling or recovery guidance. Tools will throw bare exceptions on API failures (network, auth, invalid input). Errors should be categorized (retryable vs user-fixable), include the actual invalid value and constraint, and suggest next steps. E.g., 'Invalid post ID 99999: not found. Try search-posts() to find valid post IDs.'
Parameter validation rules not documented. search-posts accepts 'title' but no minLength, maxLength, or pattern is specified. create-post title max is 70 chars (good, enforced in zod) but content is just nonempty() with no block format validation. LLMs will pass invalid block syntax and get cryptic API rejections.
No pagination support on search-posts. If WordPress returns hundreds of posts, all are included in the response, wasting tokens and risking context window overflow. Tool should accept limit and offset/cursor parameters, return a total count, and document the default/max limits.
create-post and update-post descriptions contain example values inline ('align:center', 'fontSize:large', 'borderColor:red'). LLMs often reuse example values literally in real calls, causing unintended formatting. Move examples to a resource or discovery tool, use enum constraints for valid values instead.
No permission or scope declaration. Tools accept WordPress credentials via environment variables (good for secret injection) but do not declare what permissions are required (read:posts, write:posts, etc.). This blocks least-privilege agent configuration and audit trails.