Enterprise-grade MCP Gateway that auto-discovers internal microservices (via Nacos/Eureka/etc.) and exposes their Swagger/OpenAPI endpoints as MCP tools consumable by AI clients like Cursor, Claude Desktop, etc.
MCP Gateway Enterprise exposes 4 meta-tools for service discovery and API invocation. Tool naming follows verb_noun convention (list_, search_, get_, call_) which is good. However, critical gaps undermine quality: (1) Input schemas are inferred from the description text only, no explicit JSON Schema visible in source code for any tool. (2) Parameter descriptions are minimal (10-30 chars) and lack format/constraint details. (3) Output schemas are completely undocumented, no indication of what list_services, search_api, or get_api_schema return. (4) Error handling is absent, no guidance on recovery, retryability, or actionable error messages. (5) The call_api tool accepts a generic 'arguments' object with no schema validation, inviting hallucinated payloads. This is a gateway pattern, not a production-grade tool suite.
Meta tool that invokes a backend microservice API by tool name with provided arguments
Meta tool that retrieves the full JSON Schema for a specific API tool including input and output schemas
Meta tool that lists all discovered microservices registered in the gateway
Meta tool that searches available APIs by keyword across all discovered services
No explicit JSON Schema definitions visible in source. All 4 tools lack documented input/output schemas. Schemas are inferred from brief descriptions only, violating pattern:tool and pattern:tool-description.
call_api accepts 'arguments' as a generic object with no schema validation. LLMs will hallucinate arbitrary payloads. Requires strict schema per backend API or rejection of invalid keys.
Parameter descriptions are 10-40 characters, below the 72-char baseline. 'Search keyword to find matching APIs' lacks format constraints, valid ranges, or examples of what 'keyword' means (API name? description? tag?).
No output schema documentation. LLMs cannot plan downstream calls or extract required fields. What does list_services return? Array of {name, url, version}? Pagination? Total count?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2025-06-18+ | v2 |
No error handling guidance. If call_api fails (invalid tool name, backend timeout, auth error), LLM receives no recovery hint. Missing pattern:recovery-guide and pattern:error-classification.