Multi API Web Search (MCP) - Universal Multi-Model Web Search Aggregator with Precision Model Probing, Multi-Engine Consensus, and Auto-Fallback
Two tools with identical schemas and descriptions. Both have proper input schemas with type definitions and constraints (query: string min 1, model: optional string, max_sources: number 1-30). Descriptions are present (194 chars) and explain the multi-API consensus feature. However, output schema is not documented, only inferred from code (text content + JSON payload). No parameter descriptions beyond the schema itself. Tool names lack verb clarity: 'multi_api_web_search' and 'web_search' are nouns describing the service, not action verbs. Error handling returns isError flag but lacks recovery guidance. No tool annotations (readOnlyHint, etc.). Duplicate tools (web_search as alias) violates single-responsibility principle.
Multi-API Web Search: Live internet search across your configured API models. Supports single model or multi-engine consensus ('hybrid').
Multi-API Web Search: Live internet search across your configured API models. Supports single model or multi-engine consensus ('hybrid').
Tool names lack action verbs. 'multi_api_web_search' and 'web_search' are service nouns, not action verbs like 'search_web' or 'query_web'. LLMs infer intent from verb-first naming; these names obscure the action.
Duplicate tools with identical functionality. 'web_search' is registered as an alias of 'multi_api_web_search' with the same schema and handler. This violates single-responsibility and forces LLMs to reason about which to call, wasting tokens.
Output schema not documented. Code returns {content: [{type, text}], isError?} but no formal schema is declared. LLMs cannot plan downstream operations or extract structured fields reliably.
Error messages lack recovery guidance. Errors return 'Error: {msg}' with isError=true but do not suggest next steps (e.g., 'Check API key', 'Retry with different model'). LLMs cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint. The 'READ_ONLY' risk label in metadata is not reflected in the tool definition itself.