An MCP server for UISpec that exposes catalog query and validation tools for UI component libraries
The UISpec MCP server has reasonable naming conventions and clear tool purposes, but suffers from incomplete parameter descriptions, missing output schema documentation, and lack of error handling guidance. 8 of 9 tools have descriptions, but parameter-level documentation is sparse. No visible schema documentation for outputs. Tool definitions are explicitly registered in pkg/mcp/tools.go using the mcp-go framework, but the implementation is cut off mid-definition, preventing full verification of all parameters.
Get a compact structural summary of a page's component usage for modification planning
Get full details (props, sub-components, guidelines) for one or more components
Get code examples for a specific component
Get usage guidelines. Returns global guidelines, or global + component-specific if a component name is given
Get design tokens, optionally filtered by category
Returns all component categories with their component lists
List components, optionally filtered by category and/or keyword
No output schema documentation visible. Tools return results but the response structure (fields, types, pagination) is not documented in the source. LLMs cannot plan downstream calls or extract data reliably without knowing output shape.
Parameter descriptions incomplete or missing. 'get_component_details' names parameter says nothing about expected format. 'validate_page' code parameter lacks guidance on what valid TSX looks like or what errors are caught. Baseline is 100% of A+ tools have param descriptions.
No error handling guidance. Tools do not describe error conditions or recovery steps. E.g., 'validate_page' auto_fix mode, what errors can it fix? Which are unrecoverable? What should the LLM do if validation fails?
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 46 | - | v1 |
Search components by keyword across names, descriptions, props, and sub-components
Parse TSX code, validate component usages against the catalog, and optionally auto-fix deterministic errors
Missing pagination guidance. 'list_categories' and 'list_components' could return large result sets. No mention of limit/offset parameters, result caps, or how to handle 100+ components. Baseline: 100% of A+ tools support pagination.
Tool composition risk: 'validate_page' with 'auto_fix=true' mutates state (generates fixes) but has no dry-run or confirmation step. Agents may auto-fix without understanding the changes. No support for confirmation-request pattern.
Tool names could be more specific. 'get_guidelines' is ambiguous, does it return design guidelines, usage guidelines, or both? Name like 'get_component_usage_guidelines' would be clearer.