Loading ToolBench
Sudoku100 API MCP Server - 无需注册、无需 API 密钥的数独谜题生成 MCP，支持 6 级难度和 4 种图片格式
This MCP server has critical structural and definition quality issues. The code shows three tools (generate, get_by_id, list_difficulties) but they are NOT properly registered as MCP tools. The main file (mcp/sudoku-api-mcp.js) contains a plain JavaScript object with example processing logic, not actual MCP tool registration. Tools have schemas visible in the prompt metadata, but the actual source code does not show proper MCP CallToolRequest handling or tool definition registration via the MCP protocol. Parameter descriptions exist but are generic and many are redundant across tools. No output schemas are documented. The server appears to be incomplete scaffolding rather than a functional MCP implementation.
Generate a Sudoku puzzle with optional difficulty level
Retrieve a specific Sudoku puzzle by ID
List all available difficulty levels for Sudoku puzzles
Tools are not properly registered as MCP tools. The mcp/sudoku-api-mcp.js file exports a plain JavaScript object with no MCP CallToolRequest handler, no tool definition list, and no MCP protocol compliance. This is scaffolding, not a functional MCP server.
Output schemas are completely undocumented. There is no specification of what fields the API returns, how they are structured, or what an LLM should expect. This forces LLMs to guess at response structure.
Tool descriptions are generic and do not explain WHEN to use each tool. 'Generate a Sudoku puzzle with optional difficulty level' does not explain that this is different from get_by_id (which retrieves a specific puzzle). Descriptions lack actionable context for tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
Redundant parameters across tools. All three tools accept 'action', 'difficulty', 'width', and 'format' even though get_by_id should only need 'id' and 'format'. This violates single-responsibility and creates confusion about which parameters are actually used by each tool.
No error handling or recovery guidance. No documentation of what happens when an ID is invalid (1-10000 range), what error messages look like, or how an LLM should recover from failures.
Parameter 'action' is enum-constrained but is essentially a routing mechanism that should not exist. These should be three separate tools with distinct names and parameters, not one tool with an 'action' enum. This violates the single-responsibility pattern.
Parameter descriptions are minimal. E.g., 'Difficulty level for generated Sudoku puzzle' does not explain what difficulty means, how it affects puzzle complexity, or why someone would choose 'expert' vs 'hard'. Descriptions under 50 chars typically fail to guide LLM selection.
No idempotency or safety documentation. 'generate' appears idempotent (read-only, no side effects), but this is not explicitly stated. Agents need to know which calls are safe to retry.