Demo repository containing multiple MCP server implementations and FastAPI book service
6 arithmetic tools with basic schemas and descriptions in Portuguese. Naming follows verb_noun pattern (SomarNumeros, SubtraiaNumeros, etc.). All tools have descriptions and input schemas with typed parameters and descriptions. However, descriptions are minimal (10-40 chars), parameter descriptions lack context about constraints and use cases, and output schemas are not documented. Error handling exists (e.g., division by zero check in DividaNumeros) but lacks recovery guidance. No tool annotations (readOnlyHint, etc.) despite all being read-only operations. The code is well-structured and the tools are functionally correct, but fall short of LLM-optimized documentation standards.
Divida dois números entre eles
Multiplique dois números entre eles
Eleva um número a uma potência
Extrai a raiz quadrada de um número
Adicione dois números entre eles
Subtraia dois números entre eles
Tool descriptions are too brief (10-40 chars) and lack context. 'Adicione dois números entre eles' (Add two numbers together) does not explain WHEN to use this tool vs others, WHAT the output format is, or WHAT the numeric range constraints are. Baseline: 194 chars average, 34 - 392 range for A+ tools.
Parameter descriptions are minimal ('o primeiro número', 'o segundo número') and do not specify constraints. Numeric parameters lack min/max bounds, range guidance, or use-case context. Baseline: 72 chars average for A+ param annotations; these are 15 - 20 chars.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Output schemas are not documented. The code returns string results (e.g., '(a + b).ToString()'), but the MCP tool definitions do not declare output type (string), format, or structure. LLMs cannot plan downstream operations or extract structured data without knowing the response schema.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All 6 tools are read-only and idempotent arithmetic operations, but the server does not declare these hints. LLMs cannot determine if a tool is safe for retry or speculative execution without annotations.
DividaNumeros and RaizQuadrada include error handling (division by zero, negative radicand) but error responses lack recovery guidance. 'Erro: Divisão por zero' tells the LLM an error occurred but not how to proceed. Should suggest 'Try different values' or alternative tools.
HTTP transport is correct, but no evidence of stateless request handling, _meta with logLevel, or per-request capabilities. The server uses standard .NET MCP framework (ModelContextProtocol.Server) which likely handles protocol compliance, but source does not show explicit current-spec patterns (Streamable HTTP chunking, Multi Round-Trip Requests, capability negotiation).