This is a CKAN client wrapper with 25 read-only tools. Strengths: all tools have clear descriptions (50-150 chars), complete input schemas with proper types and property descriptions, and consistent naming following verb_noun pattern (package_search, organization_show, etc.). All parameters are typed and described. Weaknesses: (1) Tool descriptions are generic and don't explain WHEN to use each tool or differentiate overlapping operations (e.g., package_list vs current_package_list_with_resources vs package_autocomplete all fetch packages but descriptions don't clarify which is best when); (2) No output schemas documented, responses from CKAN API are not shaped or constrained, forcing LLMs to guess result structure; (3) No error handling guidance, tools just wrap API responses without telling agents how to recover from 404s, timeouts, or malformed data; (4) Parameter descriptions sometimes reference internal CKAN terminology ('CKAN fq', 'CKAN facet.field') without explaining what these mean to non-CKAN users; (5) All tools are read-only (no destructive operations), which reduces risk but also means no confirmation patterns are needed; (6) No tool annotations (readOnlyHint, etc.) present to signal safety to agents. The server is functionally correct and would work, but descriptions are more reference-manual than LLM-optimized.

• Document output schemas for each tool. For package_search, specify: { packages: [{id, name, title, organization, tags, ...}], count: number, facets: {...} }. For datastore_search: { records: [{...}], total: number }. Include which fields are always present vs optional.
• Expand tool descriptions to explain WHEN to use each. Example: 'package_search: Use when you need to find datasets by keyword, organization, tag, or license. Returns paginated results with facet counts. Use package_show if you already have the dataset name or ID.'
• Add error recovery guidance to tool descriptions. Example: 'If the query returns 0 results, try package_autocomplete to discover available dataset names. If datastore_search fails with 400, the resource may not have a datastore, try resource_show to check.'
• Replace CKAN jargon with plain English in descriptions. Change 'Filter query (CKAN fq)' to 'Filters to restrict results, e.g. language:de filters to German-language datasets, organization:bundesamt-fur-statistik-bfs filters to a specific organization. See help_show("package_search") for full filter syntax.'
• Add tool annotations to all tools. Mark all read-only tools with readOnlyHint: true (all 25 are read-only). Mark idempotent tools with idempotentHint: true (most of them are, repeated calls with same args return same results).
• For package_search, convert facetField parameter to an enum or provide a discovery tool. Document valid facet fields: ['tags', 'keywords', 'organization', 'language', 'license_id']. Or add a 'get_available_facets' tool.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry