MCP server implementation running on Azure Functions, exposing tools for ping, product information, and user profile queries.
This Azure Functions-based MCP server implements 3 basic tools with acceptable naming and basic structure, but suffers from critical gaps in parameter documentation, output schema clarity, and error handling guidance. All tools follow verb_noun naming (PingFunction, ProductInfoFunction, UserProfileFunction), but descriptions are minimal (10-45 chars), parameter descriptions are trivial, and there is no documented output schema. Input validation is present but error messages are generic. The server uses Azure Functions decorators (@Function, @McpToolTrigger, @McpToolProperty) which suggest structured tool registration, but the schema definitions visible in code lack type specifications beyond string. No pagination, no error recovery guidance, no structured response schemas.
Check if the MCP server is running.
Get product information by product code.
Get basic user profile information.
Tool descriptions are extremely brief (10-45 characters) and lack guidance on when to use each tool or what they return. These are at the lower extreme with minimal context.
Parameter descriptions are trivial and under 20 characters ('Optional message to echo back.', 'The product code.', 'The user identifier.'). LLMs cannot infer parameter meaning from these minimal descriptions.
No output schema is documented anywhere. Tools return plain strings (e.g., 'Product: Laptop\nPrice: $1200\nStock: Available'), but LLMs need a structured schema to plan downstream tool calls and extract data. Unstructured text responses waste tokens and invite parsing errors.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 32 | - | v1 |
Input parameters lack type constraints and validation rules. ProductInfoFunction accepts 'productCode' but does not document valid codes (P100, P200, P300) as an enum or constrained set. Validation is inline in the function, not declared in schema. LLMs are left guessing which codes are valid.
Error messages are bare strings ('Product code cannot be null or empty.', 'Unknown product code...') with no recovery guidance. These give the agent nothing actionable.
No pagination or result limiting. ProductInfoFunction returns hardcoded results, but no documentation of limits or next_cursor behavior. If this were extended to large datasets, the lack of pagination guidance would cause context window exhaustion.
Tool composition is unclear. PingFunction is diagnostic-only ('Check if running') with no clear use case in a production agent workflow. This wastes one of three tool slots and should be removed or documented as a health-check utility only.