Four tools with complete schemas and reasonable descriptions. All tools follow verb_noun naming (list_*, get_*). Schemas include proper types, constraints (enums, min/max), and descriptions for all parameters. Output schemas are not explicitly documented in the source, which is a notable gap. Error handling is present in the API client (timeouts, retries, HTTP error parsing) but tool-level recovery guidance is absent. Pagination is well-implemented with page/per_page parameters and documented limits (1-50). No tool annotations (readOnlyHint, etc.) present. Descriptions average ~150 chars, within the 10-1024 baseline, but lack dependency hints and recovery guidance.

• Add formal output schema documentation for each tool. Example: 'Returns object with fields: data (array of vacancy objects with title, company, salary, locations, url), meta (pagination metadata with current_page, last_page, total).'
• Add tool annotations to all four tools: readOnlyHint=true and idempotentHint=true. This signals to agents that these calls are safe to retry and do not modify state.
• Enhance parameter descriptions for 'open' enum: '0 = all items (open and closed), 1 = only currently-open items (no deadline or deadline is today or later).'
• Add dependency hints to list_vacancies and list_tenders: 'To filter by category or company, you must know the valid ID. If you only have a name, the API does not provide a discovery endpoint, use the web interface to find IDs.'
• Document error scenarios in tool descriptions: 'If slug is invalid, returns 404. If rate limit is exceeded (120 requests/minute), returns 429 and retries automatically. If network timeout occurs after 120s, returns timeout error.'