Huntflow MCP has well-structured tool definitions with consistent naming (verb_noun pattern), clear descriptions in Russian, and explicit input schemas using Zod. All 14 tools are registered with descriptions and typed parameters. However, output schemas are only partially documented (12 tools have outputSchema declarations, 2 return raw text). Parameter descriptions are present but often terse (avg ~40 chars). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Error handling returns basic error messages without recovery guidance. Missing: parameter constraints (enums, min/max) on several tools, and no documentation of pagination behavior or result limits.

• Add readOnlyHint: true to all 14 tool definitions in registerStructured/registerText calls. This signals to agents that these tools are safe to call without side effects.
• Document output schemas for get_vacancy, get_applicant, and get_resume. Define a Zod schema (e.g., z.object({ id: z.number(), title: z.string(), ... })) and pass it as outputSchema parameter, even if the response is verbose.
• Enhance error messages with recovery guidance. E.g., 'Account ID 999 not found. Call list_accounts() to see available accounts.' Use the errorResult() helper to wrap actionable error text.
• Add min/max constraints to pagination parameters. E.g., count: z.number().min(1).max(100).default(30) and page: z.number().min(1).default(1). Document in descriptions: 'Page numbering starts at 1.'
• Expand parameter descriptions to 50-150 chars with context. E.g., 'account_id: The numeric ID of your HuntFlow account. Call list_accounts() to retrieve available IDs.' This guides LLMs to discover prerequisites.