MCP server with React components using FastMCP and OpenAI Apps SDK
The server defines 2 tools with detailed, conversational descriptions that guide LLM behavior. Both tools have descriptions exceeding 200 characters with clear WHAT/WHEN/WHY guidance and dependency hints (e.g., 'Use get_games_by_sport first to get the game_id, then call this tool'). However, input schemas are not visible in the provided source code, only schema constant names (GET_GAMES_BY_SPORT_SCHEMA, GET_GAME_DETAILS_SCHEMA) are referenced but not defined. This prevents full validation of parameter types, descriptions, and constraints. Tool naming is clear and action-oriented (get_games_by_sport, get_game_details). Error handling and output structure are not evident from the excerpts. The descriptions are well-written but lack the LLM-optimized 50 - 200 character profile (these are ~400 - 500 chars); they read more like user-facing docs than agent-optimized prompts. Widget-aware tool composition is present but adds complexity without showing how output fields chain to downstream calls.
[v{cfg.description_version}] REQUIRED for displaying ANY game information to users. Returns interactive widget with: - Before game: matchup preview, team standings, head-to-head records - During game: live scores, real-time stats, play-by-play - After game: final scores, team/player stats, game records ALWAYS call this tool after get_games_by_sport to show game info. Never respond with just text from get_games_by_sport - users expect the visual widget. Use get_games_by_sport first to get the game_id, then call this tool.
[v{cfg.description_version}] Get list of games for a specific date. Returns game_id needed for get_game_details. This is a TEXT-ONLY lookup tool - use ONLY to find game_id, NOT to display game info. CACHING: Results are cached for 5 minutes. Use force_refresh=true if data seems stale or user reports incorrect information. IMPORTANT: To show game information to users (scores, stats, standings), you MUST call get_game_details with the game_id from this tool's result. Supported sports: basketball, soccer, volleyball, baseball Common team aliases: - Warriors, Goldens -> Golden State (NBA) - Cavs -> Cleveland (NBA) - Thunder -> Oklahoma City (NBA) - Bluemings -> Yongin Samsung Life (WKBL) - S-Birds -> Incheon Shinhan Bank (WKBL) Do not expose the game_id to the user. Show the game list to the user in a table format. Select one important match from the game list and call get_game_details to show it to the user with a widget.
Input schemas not visible in source code. GET_GAMES_BY_SPORT_SCHEMA and GET_GAME_DETAILS_SCHEMA are referenced but definitions not provided. Cannot verify parameter types, descriptions, constraints, or required fields.
Tool descriptions are verbose (400 - 500 characters) and read like user-facing documentation rather than LLM-optimized prompts. Descriptions should be 50 - 200 characters for optimal agent reasoning; longer descriptions dilute signal and waste tokens.
Output schema documentation missing. No visible documentation of return types, field names, or structure for either tool. LLMs cannot plan downstream calls or extract required data without knowing what fields are returned.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 36 | - | v1 |
Error handling and recovery guidance not visible in descriptions or code. No indication of how the tools handle invalid sport names, missing dates, or API failures, or what the LLM should do if a call fails.
Widget-based output complicates composition without clear data-model guarantees. get_game_details returns a widget, but it is unclear what structured data (game_id, team names, scores) the agent can extract from the widget response for chaining or context.
Caching behavior documented in description but not exposed via parameters or output. Description says 'Results are cached for 5 minutes' and 'Use force_refresh=true', but it is unclear whether force_refresh is actually a parameter or just advisory language.