MCP server for Google Search with API key rotation and quota management. Supports multiple Google API keys for increased daily limits.
Single tool (google_search) with comprehensive schema and reasonable descriptions. Tool name follows verb_noun convention (search_*). Parameter schema is well-structured with types, descriptions, and constraints (enums, min/max). However, tool description lacks actionable context (WHEN to use vs alternatives, prerequisites), output schema is not formally documented, and error handling guidance is absent. The description is generic marketing copy rather than LLM-optimized prompt engineering.
Performs Google searches using the official API with automatic API key rotation. Features: - Official Google Web Search - Automatic API key rotation - Intelligent quota management - Multi-language and geolocation support Parameters: - query: Search query (required) - num: Number of results (1-10, default: 5) - start: Starting index (default: 1) - safe: SafeSearch (off/active, default: off) - lr: Language (ex: lang_fr, lang_en) - gl: Country (ex: fr, us, uk) Returns a JSON list of results with title, link, description and domain.
Tool description lacks actionable context for LLM selection. Missing WHEN to use this tool vs alternatives, prerequisites (API keys must be configured), and return value details. Current description is marketing-focused, not prompt-engineering-optimized.
Output schema is not formally documented. Parameter schema is complete, but the response structure (JSON list with title, link, description, domain fields) is only described in prose. LLMs cannot reliably extract return types from prose descriptions.
No error handling guidance. Tool description and schema lack guidance on what to do if quota is exceeded, API key is invalid, or search fails. Agents cannot self-correct without explicit recovery steps.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 8 | - | v1 |
API credentials are stored in user home directory (~/.google-search-mcp.json) after setup. While not exposed as tool parameters (good), the setup process requires manual user input and local file storage, which is fragile for agent deployments and lacks vault integration.
Parameter 'dateRestrict' format is underdocumented. Description says 'ex: d1=24h, w1=week, m1=month, y1=year' but these are example VALUES, not format constraints. Should use enums or regex pattern. LLMs will likely guess at format and pass invalid strings.
Parameter 'rights' format is underdocumented. Description mentions 'ex: cc_publicdomain' but lacks enum definition or pattern. Multiple valid values likely exist (cc_by, cc_by_sa, etc.), should be formally constrained.
No pagination guidance. 'num' parameter defaults to 5 and maxes at 10, but lacks guidance on how to fetch subsequent results beyond the 'start' index. Tool description should clarify pagination pattern (is start=1-based or 0-based? what if start > total results?).