Gold standard MCP data flow server with Sentinel architecture, dynamic configuration, and universal parsing
OmniWire-MCP has 4 tools with partial but incomplete quality. Tool names follow verb_noun convention (fetch, check, refresh, reset) which is good. However, descriptions are consistently too brief (6-33 chars, well below the 34-392 char baseline), parameter descriptions are minimal, and output schemas are not documented. Input schemas are present and use Zod validation, but descriptions lack detail about what each tool returns, when to use it, and error recovery paths. The server implements error handling in responses but does not provide actionable recovery guidance. Composition is reasonable, tools are single-responsibility, but descriptions do not explain dependencies or prerequisites.
specific diagnostic tool
Smart news fetching with filtering
Force config reload
Reset circuit breaker for a source
Tool descriptions are critically short (6 - 33 characters). Baseline is 34 - 392 chars; these fall in the bottom 10%. 'specific diagnostic tool' and 'Force config reload' lack context for LLM selection.
Output schema not documented. The tool responses return JSON-stringified content but LLMs cannot infer the field structure (title, description, publishedAt, etc. for fetch-news). Documentation must state what fields appear in each item.
Error messages lack recovery guidance. When 'No healthy sources available' or 'Source not found', the response does not suggest next steps (e.g., 'Call check-health to diagnose, then reset-source to recover').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 36 | - | v1 |
Zod schemas use .optional() and .describe() but parameter descriptions are generic or missing rationale. E.g., 'sourceId' parameter has no example format or explanation of what constitutes a valid ID.
No pagination support in fetch-news output. Results are capped at 100 items max but no offset/limit pagination or next_cursor. Large result sets could blow context windows.
fetch-news returns raw JSON stringified items. Responses should explicitly document expected fields (title, description, content, publishedAt, source) so LLMs know what to extract.