Two tools with adequate naming but significant gaps in descriptions, parameter documentation, output schemas, and error handling. Tool names follow verb_noun convention (get_alerts, get_forecast), which is positive. However, parameter descriptions are minimal, output schemas are completely undocumented, and error responses are generic strings that provide no guidance to an LLM on how to recover. The code has type hints (latitude: float, longitude: float, state: str) but the MCP schema registration is not directly visible in the provided source, only inferred from the @mcp.tool() decorator. This caps per-tool scores at 50 per HARD SCORING RULE 5. Tool descriptions themselves are brief (21 - 32 chars) but meet the 20-char floor; they lack context on when to use which tool, prerequisites, or what the output structure is.
Get weather alert for a US state
Get weather forecast for a location
Tool descriptions lack LLM-optimized context. 'Get weather alert for a US state' (28 chars) and 'Get weather forecast for a location' (35 chars) are minimal. Missing: when to use each tool, what differentiates them, what the output structure is, any rate limits or geographic constraints.
Output schema completely undocumented. Both tools return plain strings (formatted text) with no structured schema definition. LLMs cannot infer downstream tool compatibility, pagination support, or required fields for follow-up calls. Violates pattern requirement: 'Document the output schema.'
Parameter descriptions are absent or minimal. 'state' has description 'Two-letter US state code' (29 chars) but lacks format validation rules (uppercase? case-insensitive?), constraint hints, or error guidance. 'latitude' and 'longitude' lack any description in the schema, only visible in the Python docstring, which MCP clients may not parse.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 36 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 29 | - | v1 |
Error handling is generic and non-actionable. Both tools catch all exceptions with `except Exception: return None` or return bare strings like 'Unable to fetch alerts or no alerts found' or 'Unable to fetch forecast fata for this location' (note typo: 'fata'). No guidance to LLM on retry, alternative tools, or user-fixable issues. Violates recovery-guide pattern.
No pagination or result-limit documentation. get_alerts could return 100+ active weather alerts for a large state; the tool joins them into a single unstructured string, which will bloat context window and degrade LLM reasoning. Pattern requirement: 'Even if the API allows returning thousands of items, cap results at a reasonable limit (e.g. 20-50) and offer pagination.'
Tool definition registration not directly visible. Source shows @mcp.tool() decorators but no explicit schema registration with input/output types. FastMCP may infer schemas from function signatures, but this inference-based approach means per-tool scores are capped at 50 per HARD SCORING RULE. Code review cannot verify the exact schema being registered.
Typo in error message. get_forecast returns 'Unable to fetch forecast fata for this location' (should be 'data'). While minor, errors should be professional and trustworthy to agents.