The server defines 9 tools with reasonable naming convention (verb_noun pattern) but suffers from significant gaps in schema quality, output documentation, and error handling. All tools are read-only with clear intent, but descriptions are terse (10-30 chars), parameter descriptions lack format guidance, and output schemas are completely undocumented. Tools return raw JSON strings rather than structured responses, and error handling is minimal (catch-all 'Error: {str(e)}' responses). No input validation, pagination limits, or recovery guidance. This is a typical community-grade server, functional but missing production-grade polish.

• Expand tool descriptions from 10 - 30 chars to 50 - 150 chars. State WHAT the tool does, WHEN to use it (vs similar tools), and what data it returns. Example: 'Get standings for a contest by rank or user handle. Returns ranked list of participants with scores. Use after contest ends to see final results or filter specific users.' (140 chars)
• Document output schema for each tool. Define the structure of returned objects: contest object (id, name, phase, duration_seconds), user object (handle, rating, maxRating, contribution), submission object (id, creationTime, verdict). Either in tool description or as a separate schema reference.
• Add parameter constraints: contest_id must be positive integer, count must be 1 - 1000 (enforce via validation), handles must be comma-separated valid usernames (or empty). Include these in parameter descriptions.
• Replace generic error handling with categorized errors. Distinguish between: (1) API service error (timeout, 503) → suggest retry; (2) not found (404 for contest_id) → suggest get_contest_list() first; (3) invalid input (malformed handle) → suggest valid format. Return structured error objects with `error_type`, `message`, and `recovery_action`.
• Implement result limiting and pagination. Cap get_recent_submissions(count) to max 500, default 20. Return total count, next_offset for large result sets. Document in descriptions: 'Returns up to 100 submissions per call. Use offset parameter to paginate.'
• Return structured response objects instead of JSON strings. Define Pydantic models for Contest, User, Submission, Standing. Return parsed objects or dicts, not `json.dumps()` strings. This enables field extraction and chaining.

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

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