Loading ToolBench
MCP Serp has 11 tools with explicit schemas and descriptions. Strengths: consistent naming (verb-first pattern: serp_*), input parameters include enums and min/max constraints, all tools are read-only. Weaknesses: (1) Output schemas are not documented, LLMs cannot know what fields to expect from search results, limiting downstream chaining. (2) Descriptions lack strategic guidance, they explain WHAT but not WHEN to use this tool vs alternatives (e.g., when to call serp_google_search vs serp_google_images). (3) Parameter descriptions are generic; missing domain context like 'Leaving country null defaults to us' or 'Verify API key before running searches.' (4) No recovery guidance in error scenarios. (5) Information tools (list_search_types, list_countries, etc.) are discovery helpers but their descriptions don't explain when an agent should call them (e.g., 'Call this first to learn which search_type values are valid'). Average tool score is 62 across definition quality criteria.
Get a comprehensive guide for using the Google SERP tools. Provides detailed information on how to use the SERP search tools effectively, including parameters, examples, and best practices.
Search Google Images and get image results. Performs a Google Image search and returns structured image results.
Search Google Maps and get map location results. Performs a Google Maps search and returns structured map results.
Search Google News and get news article results. Performs a Google News search and returns structured news results.
Search Google Places and get local business/place results. Performs a Google Places search and returns structured place results.
Search Google and get structured results using the SERP API. Performs a Google search and returns the complete JSON response from the API, preserving all available fields and data.
Output schemas not documented. LLMs cannot know what fields (title, URL, snippet, rating, etc.) are in search results, breaking downstream tool chaining and forcing agents to make blind assumptions about response structure.
Search tool descriptions (images, news, videos, places, maps) are incomplete. They state WHAT the tool does but not WHEN to use it instead of serp_google_search. 'Performs a Google Image search...' doesn't guide the LLM, it should say 'Use when the user asks for pictures, photos, graphics, or visual content.'
Information tools (serp_list_search_types, serp_list_countries, serp_list_languages, serp_list_time_ranges) lack guidance on WHEN to call them. Descriptions should say 'Call this first to see valid values for the search_type parameter' or 'Use this to discover locale options before running a localized search.'
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
Search Google Videos and get video results. Performs a Google Video search and returns structured video results.
List commonly used country codes for Google search. Shows common country codes that can be used to localize search results.
List commonly used language codes for Google search. Shows common language codes that can be used to get results in specific languages.
List all available Google search types. Shows all available search types and their use cases. Use this to understand which search type to use for your query.
List available time range filters for Google search. Shows all time range options that can be used to filter results by date.
No error handling or recovery guidance. Descriptions do not explain what errors might occur (invalid API key, quota exhausted, malformed query) or what the LLM should do next (retry, inform user, adjust parameters). This violates the recovery-guide pattern.
Parameter descriptions lack concrete guidance. 'country code for localized results' doesn't explain what happens if null/missing, what the default is, or whether 'US' works or only 'us'. Example: 'Country code (lowercase, e.g. us, uk, cn). Defaults to us if omitted.'
The 'number' parameter note says 'More than 10 results may incur additional credits' but does not state the hard maximum or what the API returns if exceeded. Agents need clear limits to avoid surprises.