An MCP server providing tools for web scraping, weather information, news headlines, Korean baseball rankings, Google Calendar integration, quote generation, and daily briefings.
This server has significant structural and quality issues. While it uses FastMCP with HTTP transport, the tool definitions lack proper schemas, parameter descriptions are minimal or missing, and many tools have no input validation. The server mixes concerns (e.g., brief_today is a meta-instruction, not a tool), lacks error handling guidance, and several tools return raw API responses without proper structuring. The naming convention is inconsistent (snake_case is correct, but some names are overly generic). Most critically, 3 out of 7 tools have empty input schemas ({}), which violates schema documentation requirements. Output schemas are undocumented. Error responses are silent failures (e.g., 'Failed to fetch {url}' with no recovery guidance). This is a community-grade server with amateur-level tool design.
사용자의 하루 시작을 돕기 위해 날씨, 뉴스, 일정 등을 종합하여 전달합니다.
사용자에게 영감을 주는 명언을 출력합니다
한국 프로야구 구단의 랭킹을 가져옵니다
구글 RSS피드에서 최신 뉴스와 URL을 반환합니다.
도시 이름을 받아 해당 도시의 현재 날씨 정보를 반환합니다.
웹페이지의 텍스트 콘텐츠를 스크랩합니다.
Google Calendar에서 오늘 일정을 가져옵니다.
Three tools (get_news_headlines, get_kbo_rank, today_schedule, daily_quote) have EMPTY input schemas ({}). Tools must declare input parameters with proper JSON Schema.
brief_today is not a tool, it is an agent orchestration prompt. Its 'description' is 438+ chars of instructions to the agent on how to sequence other tools. This violates tool design: tools should do one thing; meta-orchestration should live in the system prompt. Either remove it and bake its logic into chat_agent.py's system_prompt, or redesign it to actually call sub-tools internally.
No tool has a documented output schema. Agents cannot plan downstream calls without knowing what fields they will receive. Tools return raw API responses (Open-Meteo JSON, Google RSS feed XML parsed to text, Daum KBO HTML response) without structure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Error handling is absent or silent. get_weather raises ValueError for unfound cities but does not document it in description. scrape_page_text returns 'Failed to fetch {url}' with no guidance on retry, fallback, or cause. No tool specifies retryable vs. fatal errors.
Security issue: today_schedule requires hardcoded 'credential.json' and 'token.json' files in the server's root directory. Credentials should be injected via environment variables or a secure vault, not read from disk. This violates the secret-injection pattern and risks credential exposure in logs.
daily_quote hardcodes model='gpt-3.5-mini' in the tool. Tool implementations should not hardcode LLM choice; this should be parameterized or injected at server startup. If the model is updated or deprecated, the tool breaks.
get_weather returns entire raw JSON from Open-Meteo API (verbose, with fields like 'weather_code', 'is_day', 'cloud_cover', etc.). Agents need a curated subset: current_temperature, weather_description, wind_speed, precipitation_chance. Unstructured output wastes tokens and increases hallucination likelihood.
get_kbo_rank returns raw HTML/JSON response text from Daum Sports API without parsing or structure. An agent cannot extract standings, win-loss records, or team names without parsing. Should return a structured list of {rank, team_name, wins, losses, games_back}.
scrape_page_text returns plain text with no metadata. Agents often need the page title, URL, domain, or excerpt for context. Should return {title, url, text, excerpt}.
Parameter descriptions are minimal or missing. E.g., get_weather accepts 'city_name' but does not document if it expects 'Seoul', 'Seoul, South Korea', or a coordinates string. scrape_page_text accepts 'url' but does not document timeout, max response size, or charset handling.