MCP server providing tools to search and retrieve Ant Design Blazor components, component information by category, and demo source code with semantic search capabilities
This server has 5 tools with READ_ONLY operations on Ant Design component documentation. Tool definitions are visible and use attributes ([McpServerTool], [Description]), but quality gaps are significant: (1) Naming is generic, 'SearchComponents', 'ListComponents', 'GetComponentsByCategory' use common verbs but lack action clarity ('Search' could mean exact match or fuzzy; naming does not distinguish from discovery tools); (2) Descriptions exist and range from 67-355 chars, exceeding the low floor but lacking LLM-optimized guidance on WHEN to use each tool or typical workflows; (3) Schemas are minimal, most tools accept only 1 - 2 string parameters with no enums, ranges, or validation hints; SearchComponentDemos is the most complex, parsing 'Component:Scenario' pairs but the input format is underscore-specified in text (not JSON Schema), forcing the LLM to guess syntax; (4) Output schemas are not formally documented, tools return markdown strings, and the structure (component name as headers, code blocks for demos) is inferred from code, not declared to the agent; (5) Error handling is basic ('Component not found', 'No demos found') with no recovery guidance ('did you mean...', 'try this other tool'); (6) No pagination, limits, or result-count documentation despite tools like ListAllDemos that could return unbounded results. Positives: all tools are READ_ONLY and safe; implementations use async patterns; demos use semantic search (embeddings) for fuzzy matching. However, per the rubric baseline (194 chars avg description, all params described, 100% of A+ tools documented output), this server falls short.
Get component information by category
List every available Ant Design Blazor demo with its component, scenario, and description. Use this command for exploration; afterwards call 'SearchComponentDemos' with 'Component:Scenario' to get the demo's source code.
List all available Ant Design Blazor components
Retrieve the full C# source code for Ant Design Blazor demo(s) by specifying component name and optional scenario. Accepts one or more 'Component:Scenario' pairs separated by commas (e.g., 'Button:Icon, Table:Editable'). Scenario can be partial or left blank to automatically choose the most relevant demo. Returns Markdown-formatted code block(s) containing the demo source.
Search for multiple Ant Design Blazor components by names
Output schemas not formally documented. Tools return markdown strings, and the structure (e.g., '### ComponentName\nDescription' headers for ListComponents, '```csharp\nSource\n```' blocks for SearchComponentDemos) is inferred from code, not declared in MCP tool metadata. Agents cannot reliably parse or extract fields without hardcoding string parsing logic.
Input parameter validation and constraints missing. SearchComponentDemos accepts 'Component:Scenario' pairs as a raw comma-separated string, but the parsing logic (split on ':', trim whitespace) is hidden in implementation. No JSON Schema format/pattern declared; LLM has no way to know the exact syntax or what happens if format is wrong. SearchComponents and GetComponentsByCategory accept bare strings with no enum constraints or length limits.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 38 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 42 | - | v1 |
Error handling lacks recovery guidance. 'Component not found' and 'Demo not found' are bare messages with no suggestion for next steps (e.g., 'Did you mean...?', 'Call ListComponents to see available options'). Per pattern:recovery-guide, errors must guide the agent on what to do next.
No pagination, limit, or result-count guidance. ListAllDemos and ListComponents offer no limit parameter and no documentation on how many results to expect. If the demo library grows, ListAllDemos could return unbounded output, exhausting context windows and degrading agent reasoning (per mxe:enforce-result-limits, cap results at 20 - 50 and document the limit).
Tool names lack clear action verbs and do not distinguish discovery from lookup. 'ListComponents' and 'GetComponentsByCategory' both list components but one filters by category. 'SearchComponents' and 'SearchComponentDemos' both search but one returns metadata, the other returns source code. Names do not clarify intent (per naming rule: 'get_channel_metadata' vs 'get_channel_messages' is clear; similar confusion exists here).
Descriptions lack LLM-optimized structure (WHAT, WHEN, HOW). Per pattern:tool-description baseline (avg 194 chars, 50 - 200 optimal), descriptions should explain what the tool does, when to call it vs. similar tools, and what it returns. Current descriptions focus on WHAT and HOW but omit WHEN and workflow guidance.