Loading ToolBench
MCP server over gold-402 curated x402 directory. Search hand-checked services and live-probe any x402 endpoint before an agent pays it.
Four tools with clear verb-noun naming and READ_ONLY annotations. All have descriptions (avg 120 chars, within baseline 34-392). Input schemas present with types and enums. However, output schemas are undocumented, responses are described in prose but not formally specified. Parameter descriptions are minimal (10-30 chars); several lack actionable constraints. Error handling is generic ('RuntimeError', 'Exception') with no recovery guidance. The server admits coverage limits and fallback behavior in instructions but tools don't expose this in responses. No pagination support despite search potentially returning many results.
Live-probe any x402 endpoint (listed or not) to verify it responds correctly to an x402 request before spending money against it.
Retrieve a single directory entry by its listing ID or name.
List all available shelves (sections) in the gold-402 directory with entry counts and descriptions.
Search the gold-402 directory for x402 services, MCP servers, SDKs, facilitators and tooling by query and/or section.
Output schemas undocumented. Tools return JSON but LLMs cannot infer field structure, types, or required fields. Responses described in prose only (e.g. 'returns payload with total_curated, matched, returned, truncated'). LLMs must guess field names and types, risking extraction errors.
No pagination support. gold402_search can return many results but accepts no limit/offset/page parameters. Large result sets blow context windows. Fallback search (line ~120) truncates silently with truncated=true but tool definition does not expose this limit or offer pagination.
Parameter descriptions too brief. 'listing' param: 'Listing ID or entry name to retrieve' (40 chars). 'url' param: 'The x402 endpoint URL to probe' (30 chars). No format constraints, no examples of valid input, no guidance on ID vs name resolution. Baseline for param descriptions is 72 chars.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Error handling lacks recovery guidance. Exceptions caught as generic 'Exception' with RuntimeError messages. No categorization (retryable vs fatal). No actionable next steps for LLM. E.g. 'Could not reach directory' does not suggest retry timing or fallback.
Fallback behavior not exposed in tool interface. gold402_search silently falls back to summary directory.json if search endpoint fails, marking result with _via and _fallback_reason. LLM cannot distinguish full-shelf results from partial summary results without parsing response metadata. Tool description does not warn of coverage limits.