xiaohongshu-mcp provides 18 tools with consistent naming (verb_noun pattern: check_, get_, delete_, publish_, list_, search_, post_, reply_, like_, favorite_), descriptions in Chinese, and detailed parameter schemas. However, significant gaps exist: (1) No documented output schemas for any tool, LLMs cannot plan downstream calls or extract chaining IDs; (2) Parameter descriptions are often verbose but lack format constraints, enums, and validation rules; (3) No error handling guidance, tools return generic responses without recovery hints; (4) Inconsistent parameter naming (xsec_token appears in many tools but is never explained as a chaining field); (5) Several tools expose internal API tokens/session fields that should be server-side injected. The server demonstrates good fundamental structure with consistent schemas and descriptions, placing it in the C+ range (fair with some good elements), but lacks the production polish of A-grade tooling.

• Add documented output schemas for all 18 tools. Each response object must declare field names, types, and whether they are chaining IDs (feed_id, user_id, xsec_token, etc.). Example: publish_content should document that it returns {post_id: string, status: string, created_at: ISO8601}.
• For tools requiring xsec_token, add explicit parameter descriptions explaining: (1) what xsec_token is (access token from prior list_feeds or search_feeds call), (2) that it must be obtained from the same feed/user before calling the dependent tool, (3) example: 'xsec_token (required): Access token returned by list_feeds or search_feeds, obtain this first and pass its value here.'
• Convert free-form string parameters to enums. Replace 'visibility' string with enum [公开可见, 仅自己可见, 仅互关好友可见]. Replace 'sort_by' and 'note_type' with explicit enum constraints in the schema (not just in descriptions).
• Add formal constraints to parameter schemas: (1) For tags arrays, add maxItems and min/max length per tag. (2) For schedule_at, add format:date-time in schema. (3) For images array, add minItems:1 and document file size/format constraints in description. (4) For limit/page parameters in list_notifications, add minimum/maximum bounds.
• Document error scenarios and recovery paths. Create a single 'Tool Error Handling & Recovery' section explaining: (1) retryable errors (network timeout, rate limit) → tell LLM to retry with backoff, (2) validation errors (invalid image format) → tell LLM to correct input and retry, (3) permission errors (unpublished account) → tell LLM action is blocked and suggest setup tools.

Inferred effective spec: 2026-07-28+.

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