Loading ToolBench
Pay-per-tool-call gating proxy for MCP servers. Wrap any MCP server with API key auth, per-tool pricing, rate limiting, and usage metering.
The paygate-mcp server defines 10 tools with consistent naming patterns (all start with 'paygate_get_' or 'paygate_list_', clear action verbs), good descriptions (all 50-180 chars), and complete input schemas with proper type definitions. All tools are READ_ONLY with no destructive operations. However, output schemas are not documented in the source, descriptions state what is returned but no formal schema definitions are provided. Parameter descriptions are good but could be more detailed on edge cases and constraints. All tools use proper verb_noun naming and are single-responsibility tools. The server demonstrates solid fundamentals but lacks documented output schemas and richer parameter constraints.
Estimate the cost of calling a tool before actually calling it. Returns expected credits and whether you have sufficient balance.
Check remaining credit balance for your API key. Returns current balance, total spent, and depletion estimate.
List active credit grants for your API key. Shows grant names, balances, expiration dates, and consumption priority.
Get server health status including uptime, backend connectivity, and system stats.
Get information about your API key including name, creation date, tags, and active status. Sensitive fields are masked.
Get detailed pricing information for a specific tool. Shows credits per call, rate limits, and any dynamic pricing rules.
Output schemas are not documented in source code. Descriptions state what is returned (e.g., 'Returns current balance, total spent, and depletion estimate') but no formal JSON Schema structures are visible for response objects. LLMs cannot plan downstream tool calls or extract specific fields without documented output schemas.
Parameter descriptions lack constraint details. 'period' enum in paygate_get_usage is declared with values but the description does not explain behavior when omitted (states 'Default: day' only). 'inputSizeBytes' in paygate_estimate_cost lacks min/max bounds and clarification on how it affects pricing. LLMs benefit from explicit numeric constraints (e.g., '1-365 days', 'positive integer up to 1MB').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | 2024-11-05+ | v1 |
Check quota usage for your API key. Returns daily and monthly call/credit usage against limits.
Check current rate limit status for your API key. Returns remaining calls, reset time, and current usage percentage.
Get usage analytics for your API key. Returns call counts, credits spent, and per-tool breakdown.
List all available gated tools with their names, descriptions, and pricing.
Error handling guidance is not visible in tool definitions. Descriptions do not explain recovery paths for common failures (e.g., 'If API key is invalid, the response will...' or 'If rate limit exceeded, retry after X seconds'). This prevents LLMs from self-correcting on failure.
No tool annotations present. Tools lack toolAnnotations (readOnlyHint, idempotentHint). While all tools are READ_ONLY by design (evident from source), this should be explicitly declared via toolAnnotations in protocol responses to help clients optimize caching and retry logic.