Loading ToolBench
Remote MCP server for verifying Korean businesses via the NTS registry
Four tools with clear, domain-specific descriptions (100-200 chars each) and well-structured input schemas. All parameters have type and description. Tool names follow verb_noun pattern (business_status, business_verify, business_batch, business_search). No output schemas documented in source. Error handling and recovery guidance absent. Security posture strong (read-only operations, no secrets in params). Composition is sound, each tool has one responsibility and batch variant exists for bulk operations.
Bulk supplier list screening: check up to 100 companies at once, $0.02 per number, with per-number results and a summary. Batch KYB screening for vendor, customer and supplier onboarding lists. Live Korean National Tax Service (NTS) data.
Search for a Korean company by name and get its business registration number. Use this when you only know what a company is called — in English or Korean — and need the 10-digit number that other lookups require. Returns ranked candidates with a confidence score and the evidence that tells similarly named companies apart: registration status, tax type and region.
Check whether a company is still operating: active, suspended or closed registration status plus tax type, looked up by business registration number. Live Korean National Tax Service (NTS) data.
KYB identity verification for due diligence and compliance: confirm that a business registration number really matches the representative (CEO) name and opening date before onboarding, contracting with or paying a counterparty. Live Korean National Tax Service (NTS) data.
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields (e.g., business_number from search results for verify). Responses must include structured field definitions.
No error recovery guidance. Tools return HTTP errors but lack actionable messages. E.g., 'Invalid business number' should suggest format (10 digits, hyphens allowed) and recovery path.
business_batch lacks per-item error reporting. If 1 of 100 numbers fails, the tool should return per-item success/failure, not a blanket error. Agents cannot retry selectively.
business_search result limit not stated. If it returns 50+ candidates, context window bloat occurs. Description should cap results and offer pagination.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
business_verify 'address' parameter marked optional but no guidance on when it improves match confidence. LLMs cannot reason about whether to include it.