MCP server for calculating various betting methods (Monte Carlo, Martingale, etc.)
Scoring was not performed
Non-action-oriented naming violates verb_noun convention. Tools are named '{method}_{action}' (e.g., 'martingale_record', 'fibonacci_status') instead of '{action}_{resource}' pattern (e.g., 'record_martingale_bet', 'get_fibonacci_session_status'). This obscures intent and forces LLMs to parse method name first, then action, backwards from the pattern that puts verb first.
Status and reset tool descriptions are generic and undifferentiated. 'Get the current status of the {Method} session' and 'Reset the current {Method} session to initial state' repeat across all 11 methods with only a placeholder name change. LLMs cannot use these to distinguish between similar tools or understand what state is returned. For a status tool, the description should explain: What fields does the response include? What state do they track? When would you call this instead of statistics?
Output schema is undocumented. The helper functions 'handleStatistics()' and 'createToolResponse()' show that responses are JSON objects with nested fields (basic.totalGames, basic.winRate, streaks.current, financial.totalWagered, etc.), but no tool description or schema export tells the LLM what to expect. LLMs cannot plan downstream reasoning or extract fields without knowing the response structure. Response schemas must be declared alongside tool definitions.
No error recovery guidance. Helper functions ('handleStatistics', 'createErrorResponse') show errors are returned as JSON with an 'error' field and 'isError' flag, but there is no LLM-facing documentation. Error responses do not include actionable steps (e.g., 'No active session. Call {method}_reset() to initialize one first' instead of just the exception message). Without recovery hints, LLMs cannot determine whether to retry, escalate, or try an alternative.
Parameter 'result' in record tools lacks sufficient description context. The enum is declared ('win' or 'loss'), but there is no explanation of what constitutes a win vs. loss, what the tool returns after recording, or whether calling it multiple times is idempotent. For a betting method tool, the description should clarify: 'Records the outcome of a single bet. Returns the next recommended bet amount. Call this after each bet resolves.'
No distinction between status and statistics tools. Both return numeric fields (win rate, ROI, streaks), but descriptions do not explain when to use status (lightweight, current session state) vs. statistics (detailed analysis including volatility and Sharpe ratio). An LLM has no reason to prefer one over the other, leading to redundant calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |