MCP server for the Autoform API from Slovensko.Digital. Provides tools for searching Slovak corporate bodies (companies, organizations) by name or registration number (IČO/CIN).
Single tool (query_corporate_bodies) with solid naming, comprehensive parameter descriptions, and well-documented input schema. Tool name starts with action verb 'query' and clearly reflects the operation. Parameter descriptions include format specifications, ranges, and examples. Output schema (SearchResult → list[CorporateBody]) is fully documented with Pydantic models. Descriptions are detailed and actionable. The server includes resources and prompts for API discovery. Main gaps: no error handling guidance in tool description (does not advise LLM what to do on rate limits, auth failure, or no results), no output pagination despite API returning up to 20 results, and descriptions could be more concise (current description is ~500 chars vs 50-200 baseline for optimal LLM consumption). Tool annotations present (readOnlyHint, openWorldHint) align with current spec.
Search Slovak corporate bodies (companies, organizations). Query format examples: - query="name:Slovenská pošta" finds companies with names starting with "Slovenská pošta" - query="cin:36631124" finds the company with IČO 36631124 - query="cin:366" finds all companies with IČO starting with "366" Returns matching records with identification numbers (IČO, DIČ, IČ DPH), addresses, establishment dates, and links to additional registry data.
No error handling guidance in tool description. LLM does not know what to do if the query returns 0 results, API returns 401, or rate limits are hit.
Tool description exceeds optimal length (~500 chars). Includes format examples inline rather than relying on parameter descriptions.
Output pagination not implemented. Although limit caps at 20, no offset/cursor mechanism or next_token provided for clients to fetch additional results. Per pattern, even small result sets should include pagination when API supports iteration.
Token injection via environment variable and HTTP headers works, but no audit logging of tool calls. No trace showing which user/agent called the tool, with what parameters, and what the result was. Per pattern, agent actions must be traceable for compliance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 62 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | - | v1 |