A simple math operations MCP server using FastMCP
This is a minimal math utility server with 4 simple tools. All tools are properly registered via @mcp.tool() decorator with explicit docstrings and type hints. Input schemas are correctly inferred from function signatures (int parameters with descriptions). However, the descriptions are extremely terse (10-30 chars), well below the 50-200 char baseline for LLM-optimized descriptions. No output schemas are documented. Error handling exists (divide checks for zero), but error messages lack recovery guidance. No tool annotations, no parameter constraints, and the divide tool has a type mismatch (accepts int but returns float) that isn't reflected in schema documentation. The tools follow single-responsibility (one operation each) and are composable, but lack the context and guidance LLMs need for confident selection and error recovery.
Add two numbers
Divide two numbers
Multiply two numbers
Subtract two numbers
Descriptions are under 20 characters for all tools (10-30 chars observed). Below baseline of 50-200 chars for LLM-optimized descriptions.
No output schemas documented. LLMs cannot plan downstream tool calls or extract results. Baseline: 100% of A+ tools have documented return types.
divide() accepts int inputs but returns float. Schema shows input as int but return type mismatch is not documented. This can confuse agents about expected output type.
Error handling in divide() raises ValueError('Cannot divide by zero') with no recovery guidance. Per pattern, errors should tell LLM what to do next.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 46 | - | v1 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). While not required, current spec rewards their presence for clarity.