A Flask-based HTTP API server with a Telegram bot client that provides coin flip and echo functionality. Not an MCP server - this is a standard REST API with a Telegram bot consumer.
This MCP server has critical definition quality gaps. Four tools are defined, but descriptions are minimal (10 - 20 chars), parameters lack type information and descriptions, and output schemas are entirely undocumented. The server is NOT an MCP server at all, it is a Flask HTTP API that does NOT implement the Model Context Protocol. No MCP handshake, no tool registration via MCP protocol, no schema-based parameter validation, and no structured error handling. The bot/bot.py makes HTTP calls to the API, but this is a client using REST, not an MCP transport. The codebase has zero MCP server scaffolding (no mcp.server.Server, no Tool model, no initialization, no capability negotiation). This appears to be a misclassified submission, a Flask REST API, not an MCP server. Scoring it as if it were an MCP server reveals the quality baseline: tool descriptions are too brief, schemas are inferred only from the Flask route implementations, parameters are unmapped to formal definitions, and outputs are undocumented.
Принимает JSON и возвращает его обратно
Возвращает случайный результат подбрасывания монетки
Проверка здоровья API
Возвращает статус API с временной меткой
Not an MCP server. The codebase is a Flask HTTP REST API with a separate Telegram bot client. No MCP protocol implementation, no server scaffolding, no Tool registration, no capability negotiation. This is a misclassified submission.
All tool descriptions are trivially short (10 - 20 chars). Descriptions lack WHAT the tool does, WHEN to use it, and prerequisites. Examples: ping='Возвращает статус API с временной меткой' (40 chars, but vague), flip='Возвращает случайный результат подбрасывания монетки' (54 chars, acceptable), echo='Принимает JSON и возвращает его обратно' (41 chars), health='Проверка здоровья API' (21 chars, too brief). None answer all three questions or guide LLM selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 18 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Input/output schemas are completely absent. Tools have no formal JSON Schema definitions, no parameter type declarations, no required fields, no field descriptions. Only 'echo' has a partial schema in the metadata (type=object, properties with 'data'), but it lacks descriptions for the 'data' parameter itself. Baselines: 100% of A+ tools have documented return types; 100% of A+ tool params have descriptions.
No parameter descriptions. 'ping' has no parameters (GET only) but no output schema. 'flip' has no parameters. 'echo' accepts 'data' (object) but the field is undescribed, LLMs cannot know what JSON structure is expected, leading to malformed calls. 'health' has no parameters.
No output schema documentation. Flask routes return JSON objects (e.g., ping returns {ok: true, ts: int}, flip returns {result: 'heads'|'tails'}, echo returns the input JSON, health returns {status: 'healthy'}), but no tool declares what fields clients should expect. LLMs cannot plan downstream tool chains or extract fields reliably.
Error handling is minimal and not LLM-friendly. Flask routes catch exceptions generically (e.g., echo catches Exception and returns {error: str(e)}, health returns only {status: 'healthy'}). Errors do not guide recovery: no categorization (retryable vs fatal), no actionable next steps, no constraint violations. A 500 error on 'echo' with no JSON gives a stack trace, not 'Send a JSON object with a "data" field.'
Tool naming is inconsistent with verb_noun convention. 'ping', 'flip', 'echo', 'health' are nouns or bare commands. Better names: 'get_api_status', 'flip_coin', 'echo_data', 'check_health'. Current names are not action-driven, LLMs infer intent less clearly. Baseline: 90% of A+ tools start with action verbs (get, list, create, search, update, send, delete).
No error categorization or recovery paths. If 'echo' is called with invalid JSON or missing 'data' field, the error does not tell LLM whether to retry, ask user, or abandon. No dry-run or confirmation for 'flip' (harmless) or other tools. No rate-limiting guidance.