MCP server that provides book recommendations based on user input using the Google Books API and finds nearby bookstores using the Google Maps API
This server exhibits significant quality gaps across naming, descriptions, and schemas. Tool names lack consistent verb prefixes (recommend_books has no clear action verb). Descriptions are verbose and include implementation details (API URLs, example queries) rather than LLM-optimized guidance. Input schemas are present but lack proper typing and constraints. The server relies on stateful caching (cache dict) which violates MCP's stateless request model. Error handling is minimal, no recovery guidance, no categorization of error types. The three tools have moderate documentation but fail to meet production baselines for clarity and completeness.
Uses the google maps api to find bookstores based on the users location input, City,State should return up to 5 nearby stores using the places api as it allows searching based on queries https://developers.google.com/maps/documentation/places/web-service/text-search url = "https://places.googleapis.com/v1/places:searchText"
Using information from the official site https://developers.google.com/books/docs/v1/using Here is an example of searching for Daniel Keyes' "Flowers for Algernon": GET https://www.googleapis.com/books/v1/volumes?q=flowers+inauthor:keyes&key=yourAPIKey switching to author and title search, genre can be extracted by the llm for recommendation purposes
Recommend books using ONLY the cached located book.
Stateful caching violates MCP stateless request model. recommend_books relies on cache['last_located_book'] set by locate_book, creating hidden state dependencies.
Tool descriptions include implementation details (API URLs, example payloads, JSON structure) instead of LLM-optimized WHAT/WHEN/WHY guidance. Descriptions mention 'https://www.googleapis.com/books/v1/volumes' and 'https://places.googleapis.com/v1/places:searchText', which clutter the prompt and invite literal reuse by LLMs.
No output schemas documented. LLM cannot infer what fields to expect from each tool's return value, preventing downstream tool chaining and forcing LLM to guess field names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 36 | - | v1 |
recommend_books has empty input schema (no parameters). This is unusual and forces hidden reliance on global cache state rather than explicit input. Tool should accept book_genre or similar as input.
Parameter 'location' in find_bookstores lacks format constraints. Description says 'City, State format' but no validation, enum, or regex pattern is enforced. LLM may pass invalid formats.
Error handling returns plain text strings with no recovery guidance. E.g., 'No books found matching the title and author, please try again' does not tell LLM WHAT to try or which tool to call next.
API keys exposed: GOOGLE_MAPS_API_KEY is read from environment in find_bookstores tool, but locate_book does not specify how API keys are managed (no key parameter visible, assumes implicit injection). Inconsistent secret handling.
Tool composition issue: recommend_books assumes locate_book was called first and cached the result. If client session is stateless (as MCP specifies), this chain fails silently. Tools are not properly composable.