Loading ToolBench
An AI gateway proxy router that handles request routing, PII sanitization, security policy enforcement, and token cost tracking for AI model requests
This server has critical foundational deficits. The single tool 'route_gateway_request' has a reasonable name but lacks essential schema documentation, parameter descriptions are minimal, output schema is completely undocumented, and there is no error handling guidance. The codebase provides only a stub implementation with no actual routing logic, PII sanitization, cost tracking, or security policy enforcement. The tool definition exists but parameter and output documentation are insufficient for LLM-driven reliable use. This is a proof-of-concept, not a production-grade MCP server.
Routes incoming AI gateway requests through security policies, sanitizes PII, and tracks token costs
Output schema completely undocumented. Tool returns a dict with fields 'forwarded_status', 'pii_sanitized', 'token_cost_usd' but LLMs have no schema to understand the return structure, types, or possible values. Cannot plan downstream tool calls or extract data reliably.
Input parameters lack type information and minimal descriptions. 'incoming_request_json' is described as 'The incoming request payload to be routed' (41 chars), does not explain format, structure, required fields, or examples. 'security_policies' description is 115 chars but lacks enum constraint for valid policy names; states 'e.g., PII_MASK, RATE_LIMIT' instead of declaring an enum. This invites hallucinated policy values.
No error handling guidance. Stub implementation always returns success with mock data. No documentation of failure modes (e.g., invalid request format, unsupported policy, gateway unavailable) or recovery actions. LLMs cannot distinguish real errors from mock responses or know what to do on failure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No parameter validation or security constraints documented. 'incoming_request_json' accepts any dict with no validation. No mention of PII detection scope, rate limit enforcement, token cost calculation methodology, or how security policies are applied. Tool description claims 'sanitizes PII' but implementation does not, misleading documentation.
Stub implementation. The tool returns hardcoded mock data (forwarded_status='FORWARDED_200_OK', pii_sanitized=True, token_cost_usd=0.00042) regardless of input. No actual request routing, PII masking, rate limiting, token tracking, or security policy enforcement. This is not functional for production use.
security_policies parameter default is None but description says 'Optional list', inconsistent. No guidance on behavior when policies is None, empty list, or contains unknown values. LLMs will not know what happens when the parameter is omitted.