MCP server for Basecoat UI components - provides HTML components and usage examples
This MCP server provides 7 read-only tools for accessing Basecoat UI components. All tools are explicitly registered with schemas and descriptions. Naming follows verb_noun convention consistently (get_*, list_*, search_). However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions are minimal (under 20 chars for most parameters). Output schemas are not documented, the server returns responses but LLMs cannot plan downstream operations without seeing what fields to expect. Error handling is basic; no recovery guidance. The 'get_component' and 'get_usage' tools accept free-form strings without validation or enum constraints, inviting hallucinated values. No tool annotations (readOnlyHint) despite all being read-only. Composition is reasonable, each tool has a single clear responsibility, but search and category tools overlap without differentiation.
Get all components in a specific category
Get HTML code for a specific Basecoat component
Get Basecoat CSS setup code with CDN links
Get theme switcher script for dark/light mode
Get usage documentation for a component
List all available Basecoat components organized by category
Search for components by name or category
Output schemas are not documented. LLMs cannot see what fields tools return, preventing them from chaining operations or extracting relevant data. For example, get_component returns {name, category, html, file} but this structure is not declared in the tool schema.
Parameters lack enum constraints and validation guidance. 'component' in get_usage and 'name' in get_component accept free-form strings with no enumerated valid values, inviting hallucinated component names. The server silently returns 'Component not found' on invalid input, offering no recovery hint.
Parameter descriptions are minimal (under 20 chars in most cases). 'component' is described as 'Component name (e.g., "button", "card", "input")', example values are present but not formal constraints. 'category' parameter in get_category lacks any description of valid category values; LLM must infer from context.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 50 | - | v1 |
Tool descriptions lack LLM-optimization guidance. Descriptions do not answer WHEN to use this tool vs alternatives (e.g., when to use search_components vs get_category?). No dependency hints provided (e.g., 'Call list_components first if you don't know the exact name'). Descriptions are under 100 chars, below the 194-char baseline for A-grade tools.
Error responses provide no recovery guidance. When 'Component not found' or 'Category not found', the error message does not suggest next steps (e.g., 'Try search_components() with a partial name'). This forces the LLM to guess at recovery.
No tool annotations present. All 7 tools are read-only (no side effects), but none are marked with readOnlyHint:true. This forces LLMs to reason about safety and retry semantics without explicit signal.
search_components and get_category tools overlap in functionality without clear differentiation. Both search/filter the component list. LLM reasoning is wasted deciding between them. Consider consolidating or clarifying when to use each.