A dining assistant with access to BrightData MCP tools for searching restaurants, scraping menus, and creating dining plans with bill estimates
This MCP server exposes 3 tools via FastAPI/HTTP. All tools have descriptions and basic input schemas, but several critical quality gaps prevent a higher score. Tool names are verb-based (search_engine, scrape_as_markdown, create_dining_plan), which is good. However, parameter descriptions are minimal or absent, schemas lack depth, and error handling guidance is not visible in the code provided. The tools appear to wrap external APIs (Google/Bing search, web scraping, restaurant planning) but the server code snippet cuts off before showing complete tool implementations, error handling, or response shaping. Based on visible code in app.py, tool definitions are present but incomplete.
Create a dining plan with menu items and bill estimate for a restaurant.
Scrape a webpage and return content as markdown.
Search for information using Google, Bing, or Yandex search engines.
Parameter descriptions are minimal or absent. 'engine' param in search_engine has a 1-line description but no guidance on consequences of each choice (Google returns X, Bing returns Y). Parameter 'url' in scrape_as_markdown has no description. 'restaurant_url' in create_dining_plan has generic description with default value but no guidance on format or required structure.
Input schemas lack detail. 'search_engine' accepts 'engine' with a string type and default 'google', but no enum constraint listing valid values (google, bing, yandex). LLMs may hallucinate invalid engine names. 'scrape_as_markdown' URL param has no format constraint (http/https only? URL validation?). 'create_dining_plan' restaurant_url has empty default string, but no guidance on when it is optional vs required or what happens if omitted.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Output schemas are not documented. Code shows tool definitions but no specification of what each tool returns. search_engine likely returns search results, but structure (array of results? JSON object? markdown text?) is not stated. scrape_as_markdown presumably returns markdown content, but max length, truncation behavior, and field structure are undocumented. create_dining_plan output format (menu items array? bill estimate object?) is not visible. LLMs cannot plan downstream calls without knowing return structure.
Error handling is not visible or documented. Code shows basic try-catch in search_engine with generic error message 'Search failed: {e}', which does not guide LLM recovery. No indication of retryable vs fatal errors, rate limits, timeout handling, or suggestions for user-fixable issues. For a tool calling external APIs (Google, web scrape, restaurant data), robust error guidance is critical.
Tool descriptions are under 100 characters and lack detail on when to use each tool vs alternatives. 'Search for information using Google, Bing, or Yandex search engines' (72 chars) does not explain: Which engine is best for restaurant data? Does Yandex cover US restaurants? When should the agent prefer search vs scrape? Baseline for A+ tools is 194 chars avg with clear use-case and prerequisite info.
create_dining_plan tool name is ambiguous. Does it create a reservation? A menu selection? A bill estimate? Or all three? The name suggests a single output 'plan', but no description clarifies. 'Create a dining plan with menu items and bill estimate' (68 chars) is too short; a production tool would explain: what is a dining plan? Does it require a reservation? What data is needed vs optional? Does it call search_engine and scrape_as_markdown internally or expect the agent to pre-fetch data?
No idempotency or confirmation patterns visible. create_dining_plan appears to modify state (create a plan, possibly reserve a table), but no indication of whether repeated calls are safe, whether a dry-run mode exists, or whether confirmation is required before booking. Agents making mistakes could trigger unintended reservations.
No pagination or result limits documented. search_engine calling external APIs could return hundreds of results. No indication of max_results parameter, pagination support, or limit enforcement. Baseline pattern requires list tools to accept limit and offset/cursor params.