News MCP server providing access to comprehensive market and financial news with advanced filtering options through the MarketAux API.
Single-tool server with a moderately strong tool definition. The market_aux_news_search tool has a well-structured Zod schema with all 8 parameters described, proper typing, and a substantive description (195 chars). However, there are several significant gaps: (1) no output schema documentation, executeSearchAll returns generic text without specifying the structure agents should expect; (2) parameter descriptions lack formatting guidance (e.g., what format for dates? 'YYYY-MM-DD'?); (3) no error handling documentation, the tool catches errors and returns text, but does not guide the agent on recovery strategies; (4) no pagination/limit guidance despite the tool likely returning potentially large result sets; (5) missing tool annotations for security/idempotency. The naming 'market_aux_news_search' is verb-noun compliant. Overall, the tool is above average for community servers but lacks production-grade refinement in output documentation and error guidance.
Search comprehensive market and financial news with advanced filtering options. Find news articles by stock symbols, entity types, industries, countries, languages, and date ranges. Perfect for market research, investment analysis, and staying updated on specific financial entities or market sectors.
Output schema not documented. executeSearchAll returns JSON text without declaring field structure, types, or pagination metadata. Agents cannot plan downstream operations or know whether to expect a list, object, or metadata fields.
Date parameter formats not specified. published_before, published_after, and published_on accept strings but lack guidance on format (ISO 8601? MM/DD/YYYY? Unix timestamp?). This invites LLM hallucination of invalid date formats.
No pagination or result limit documentation. The tool queries an external API but does not declare a limit on returned results or support for pagination. Large result sets will blow context windows and waste tokens.
Error responses are generic text without recovery guidance. When the API fails, the tool returns 'Error searching news: <message>' but does not advise the agent whether to retry, adjust parameters, or escalate. Pattern: recovery-guide.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 38 | - | v1 |
No tool annotations for idempotency or destructiveness. The tool is READ_ONLY per metadata, but this is not declared via toolAnnotations (readOnlyHint). Agents cannot infer safe retry behavior from the tool definition alone.
Language parameter ambiguous. Accepts 'comma separated list of languages' (e.g., 'en,fr') but does not specify language code format (ISO 639-1? ISO 639-3?). Agents must guess or experiment.