MCP server for ARCA/AFIP web services documentation — hybrid semantic + BM25 search over 50+ official Argentine tax-agency web services.
Four well-named, read-only tools with complete Zod schemas and detailed descriptions. All tools follow verb_noun naming (search_, list_, get_). Descriptions are comprehensive (150 - 250 chars) and include usage guidance. Schemas are properly typed with min/max constraints. However, output schemas are not formally documented, only inferred from code. Error handling is present but minimal (no recovery guidance or actionable next steps). No tool annotations (readOnlyHint, etc.) despite all being read-only. Security is solid (no secrets in params, input validation via Zod), but descriptions could better explain when to call each tool vs. alternatives.
SEGUNDO PASO cuando search_docs no trae suficiente detalle. Devuelve el manual PDF completo reconstruido (hasta ~60k chars, equivalente a ~15k tokens). USÁ ESTO en vez de fallback a fetch/WebFetch del PDF directo cuando necesites: schema XML completo de response/request, tabla de códigos de error entera, lista completa de métodos de un WS, o cualquier detalle no-truncado. Input: la URL exacta que aparece como `source` en los resultados de search_docs.
Devuelve metadatos y tabla de contenidos de un web service específico: nombre completo, URLs oficiales y secciones indexadas.
Lista todos los web services AFIP/ARCA indexados con su ID y descripción. Usá este tool primero para saber qué wsId pasarle a search_docs o get_web_service.
PRIMER PASO para cualquier consulta sobre AFIP/ARCA. Búsqueda semántica que devuelve 8 fragmentos relevantes con su URL fuente. Ideal para localizar: códigos de error, nombres de campos, reglas de validación, ejemplos. Si los fragmentos están truncados o no traen el schema/respuesta completa, LLAMÁ A get_document con la URL fuente que aparece en los resultados.
Output schemas not formally documented. Tool descriptions explain what is returned (e.g., 'devuelve 8 fragmentos relevantes'), but JSON Schema output types are absent. LLMs cannot reliably parse or chain results without explicit output schema definitions.
Error responses lack recovery guidance. When a document is not found (get_document), the response says 'no indexado' but does not suggest calling search_docs or list_web_services. Errors should guide the LLM to the next action.
No tool annotations despite all tools being read-only. Adding readOnlyHint=true to all four tools would signal to agents that these are safe to call without side effects, enabling more aggressive caching and retry strategies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
list_web_services lacks pagination and result limits. If 50+ web services are indexed, returning all at once wastes tokens. Should accept limit/offset params and return total count.
search_docs topK default is 8 but description does not explain why 8 is optimal or when to increase it. For large result sets, LLMs may need guidance on result limits vs. token cost.