A centralized MCP gateway that proxies tool calls to backend MCP servers with policy enforcement, redaction, audit logging, and telemetry
This server exposes only 2 test fixture tools with minimal production quality. Both tools have extremely short descriptions (under 40 chars), lack parameter descriptions entirely, and one tool (delete-everything) has an empty input schema with no documentation of its destructive behavior. The echo tool has a basic schema but no parameter description for 'text'. Neither tool follows verb-noun naming conventions clearly ('echo' is ambiguous; 'delete-everything' lacks clarity on scope). No output schemas are documented. Error handling is absent, no guidance on what happens on failure or how to recover. This is a test fixture, not a production-grade tool suite.
dangerous stand-in tool
echoes arguments back
delete-everything has empty input schema (no properties) and no description of what it deletes or its scope. Destructive tools must document exactly what data is affected and require confirmation.
No parameter descriptions. The 'text' parameter in echo has no description explaining what it accepts, expected format, or constraints. LLMs cannot infer parameter meaning from names alone.
Tool descriptions are under 40 characters and lack context. 'echoes arguments back' and 'dangerous stand-in tool' do not explain when to use each tool, what they return, or their purpose in the system.
No output schemas documented. LLMs cannot plan downstream calls or extract return values without knowing the structure of responses (e.g., does echo return {echoed: string} or {text: string}?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 30 | 2026-07-28+ | v2 |
No error handling guidance. If echo receives null or delete-everything fails, there is no recovery path or actionable error message for the LLM.