MCP Server for multi-tenant token cost tracking, rate limiting (RPM/TPM), and budget quota management for LLM API calls
Two tools with basic schemas and minimal descriptions. check_rate_limit lacks parameter descriptions entirely (tenant_id, max_rpm, max_tpm have no descriptions in the schema). record_usage similarly lacks parameter descriptions. Both tool descriptions are under 100 chars and lack context on when to use them, prerequisites, or what the LLM should do with the response. No output schema documented. No error handling guidance. No parameter constraints (enums, ranges). Naming is verb-noun compliant but generic. No idempotency hints or state-modification warnings despite record_usage being a WRITE operation.
Evaluate sliding-window RPM and TPM quota.
Record LLM token consumption and check spending limits.
Parameter descriptions missing entirely. Schema defines tenant_id, max_rpm, max_tpm for check_rate_limit and tenant_id, model, prompt_tokens, completion_tokens for record_usage, but none have descriptions in the inputSchema. LLMs cannot infer parameter meaning from names alone.
No output schema documented. Responses are JSON-serialized text blobs. LLMs cannot plan downstream calls or extract structured data without knowing the response shape (fields, types, required vs optional).
Tool descriptions too brief and lack context. 'Evaluate sliding-window RPM and TPM quota' (54 chars) and 'Record LLM token consumption and check spending limits' (56 chars) do not explain WHEN to call, WHAT the response means, or HOW to use it. Baseline is 194 chars average for A+ tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
No parameter constraints. max_rpm and max_tpm accept unbounded integers with no min/max guidance. LLMs could pass absurd values (0, -1, 999999999). No enums for model selection, free-form string invites hallucinated model names.
record_usage marked WRITE but description does not warn of side effects or idempotency. LLMs do not know if retrying the same call twice will double-charge the tenant. No idempotentHint or destructiveHint annotation.