MCP server that exposes tools for weather data, stock market data, and web search via real external APIs
Server defines 3 tools with explicit schemas and descriptions visible in server/main.py. All tools have names following verb_noun convention (get_*, web_search). Descriptions are present but minimal (16-48 chars). Input schemas are properly structured with type and required fields. However, schemas lack depth: no enums where appropriate (e.g., web_search.num_results should have min/max constraints), no documented output schemas, no error recovery guidance. Tools are read-only with no state-modifying consequences, reducing some risk. Parameter descriptions are sparse, only 'City name' for get_weather.city, '(e.g., Bangalore, London)' which violates the rubric's prohibition on example values in descriptions. Overall coherent but minimal quality, typical of early-stage showcase servers.
Get real stock market data for a given stock symbol.
Get real-time weather data for a given city.
Search the internet using Google Custom Search.
Descriptions contain example values instead of formal constraints. 'City name (e.g., Bangalore, London)' and 'Stock symbol (e.g., AAPL, TSLA)' train LLMs to reuse examples literally rather than inferring intent. Should use enum or pattern constraints instead.
No output schemas documented. Callers cannot anticipate what fields the tools return (e.g., does get_weather return 'temp' or 'temperature'?), forcing trial-and-error discovery and downstream tool failures.
Parameter descriptions are minimal (1-3 words). 'City name' lacks context: Is it required? What format? Fuzzy match or exact? Should describe format, range, and when to use.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
web_search.num_results parameter has no min/max constraints. Unbounded integers let LLMs pass absurd values. Should specify range (1-10) and document default behavior.
Error handling is generic. call_tool() catches all exceptions and returns 'status: error' with minimal context. No recovery guidance (e.g., 'Invalid city. Try searching for the city name first').
Tool descriptions do not explain WHEN to use each tool or distinguish from alternatives. 'Get real-time weather data for a given city' is functional but lacks discovery hints. Should answer: What does it return? When should the LLM call it?