Model Context Protocol (MCP) server implemented as an Azure Function with isolated worker model
Critical gaps across all dimensions. Tool schemas are present but lack proper JSON Schema structure with type definitions, parameters are documented as strings in a Dictionary rather than formal schema objects. Descriptions exist but are generic and lack actionable guidance. No error handling patterns, no parameter constraints (enums, ranges), no output schema documentation. Tool names lack clear verb-noun structure (e.g., 'triggerLogicApp' uses camelCase instead of snake_case convention). No evidence of idempotency, permission gates, or security considerations. The server exposes internal API paths and workflow names as parameters without sanitization guidance.
Calls an internal API endpoint
Retrieves business data including customer counts, sales data, and top products
Returns the OpenAPI specification for this MCP server (MCP standard operation)
Returns a list of all available operations and their descriptions (MCP standard operation)
Triggers a specified Logic App workflow
Input schemas lack JSON Schema type definitions. Parameters are stored as Dictionary<string, object> with string descriptions only. No type, format, enum, or constraint information is machine-parseable. This violates the constrained-input pattern and forces LLMs to guess valid values.
No output schema documentation. Tools return MCPResponse objects but the structure of the 'result' field is not documented. LLMs cannot plan downstream tool calls or extract required fields without knowing what data is returned.
Tool names use camelCase (triggerLogicApp, callInternalApi) instead of snake_case verb_noun convention (trigger_logic_app, call_internal_api). This deviates from production baseline where 90% of A+ tools use snake_case and start with action verbs.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | <=2025-11-25 | v2 |
No error handling guidance. Error responses contain only a message string. No categorization of errors as retryable, user-fixable, or fatal. No actionable recovery hints (e.g., 'Try search_users() first'). This violates the recovery-guide pattern.
callInternalApi exposes internal API paths and HTTP methods as parameters without validation or sanitization guidance. No protection against path traversal, command injection, or unintended API calls. Missing security pattern: tool-gateway.