Nordic Data MCP exposes 10 tools with consistent naming (verb_noun pattern) and basic descriptions. However, critical gaps undermine production readiness: (1) Parameter descriptions lack format constraints, ranges, and validation guidance, e.g., 'country' accepts 'lowercase ISO code' but no enum or pattern is enforced; (2) Output schemas are not documented, LLMs cannot predict response structure for chaining; (3) Error handling is absent, no recovery guidance or categorization; (4) Three meta-tools (list_endpoints, get_endpoint_schema, call_endpoint) expose a generic API gateway pattern that invites hallucinated endpoint names and parameter misuse. Tool descriptions average ~80 chars (below 194-char baseline), lacking WHEN/WHY context. No tool declares permissions, audit trails, or idempotency guarantees.

• Document output schemas for all 7 domain tools (lookup_company, kyb_full, screen_sanctions, validate_vat, company_enriched, autocomplete_address, lookup_lei). Include field names, types, and which fields are required for chaining (e.g., company_id, person_id).
• Replace free-form 'country' parameter with enum constraint listing all supported ISO codes (DK, NO, SE, FI, FR, etc.). Add similar enums for validate_vat's country parameter (list valid VIES codes).
• Add pagination to screen_sanctions and autocomplete_address: accept limit (1 - 100, default 20) and offset/cursor parameters. Document max result count in tool description.
• Deprecate or heavily restrict call_endpoint. If kept, whitelist allowed endpoints and validate parameters against a schema registry. Otherwise, replace with explicit domain tools only.
• Add error recovery hints to all tools: 'If company not found, try search_company() with partial name' or 'Invalid VAT format, ensure country code matches VIES requirements.'