Loading ToolBench
MCP server that provides a tool-calling interface for fetching news articles from a MongoDB database, designed to work with ChatGPT's tool-calling feature
Single tool server with clear implementation but significant documentation and design gaps. The get_news tool has a properly defined schema with type constraints and descriptions, but lacks critical production-readiness features like error recovery guidance, output schema documentation, and security considerations. Descriptions are adequate but not LLM-optimized. The tool follows verb_noun naming (get_news) correctly. Overall fair/poor, many community servers land here.
Fetch news articles from MongoDB database. This tool queries a MongoDB collection for news articles and returns them in a structured JSON format. It supports filtering by topic, limiting results, and sorting by publication date.
Output schema not documented. Tool description states 'returns them in a structured JSON format' but does not specify the exact fields, types, or structure the LLM should expect. This forces the LLM to infer output shape and makes downstream tool chaining difficult.
Error handling provides no recovery guidance. If MongoDB connection fails or a query returns empty results, the tool returns an empty list or generic error. The LLM receives no actionable message about what went wrong (network issue? invalid topic?) or what to try next.
Parameter descriptions include example values (e.g. 'Examples: AI, finance, technology, sports'). LLMs tend to reuse example values literally rather than generalizing to user context, potentially ignoring valid topics not listed.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 56 | - | v1 |
No result limits documented in tool description. Implementation caps limit at 50, but the description does not state this. Without an explicit limit, LLMs may assume unbounded results. Clarify 'Returns up to 50 articles per query' in the description.
No pagination support. Tool accepts a limit but no offset/cursor for fetching beyond the first N results. For a news service, users often want to browse beyond the top 5 articles, but there is no way to do so without refetching and losing state.
Missing tool annotations. Tool is read-only but does not include readOnlyHint annotation in schema. This prevents MCP clients from optimizing caching or understanding the tool's safety profile.