MCP server connecting AI agents to Scavio: web search, page extraction, and structured e-commerce, social, travel, jobs, real-estate, app-store, ad-library and company-filing data. 191 tools across 31 platforms plus Extract, one API key.
Scavio MCP server demonstrates solid definition quality across 16 e-commerce and travel tools. All tools have descriptions and visible input schemas with type definitions. Tool names follow verb_noun convention (search_*, get_*). Descriptions are concrete and include critical operational details (pagination limits, credit costs, parameter constraints). However, several patterns from the 54 Agentic Tool Patterns are underutilized: output schemas are not formally documented in the provided source, tool annotations (readOnlyHint) are absent despite all tools being read-only, and some parameter relationships lack explicit documentation. Error handling via handleApiError() is present but focused on HTTP status codes rather than recovery guidance. The server represents above-average quality but falls short of A-grade (80+) due to incomplete schema documentation and missing tool annotations.
Get full Airbnb listing details, amenities, host, rules, photos, rating breakdown. NO PRICES here; use search_airbnb for pricing. 1 credit.
Get Airbnb review text for a listing. Rating breakdown is on get_airbnb_listing, not here. Paginate with limit+offset. Send limit explicitly or upstream returns only 7. 1 credit/page.
All sellers for an Amazon ASIN with prices and conditions. First page only, no pagination. 1 credit.
List supported Amazon marketplaces and country codes. Free, no args.
Full Amazon product page for an ASIN. Price is buy-box price; use get_amazon_offers for competing sellers. 1 credit.
Get full App Store listing. Accepts numeric id or bundle id (e.g. 'notion.id'). Full URL rejected (extract the id). Unknown id = billed 404. 1 credit.
Output schemas not formally documented. Tools return results but the source code does not explicitly declare what fields, types, and structures agents should expect in responses. This forces LLMs to infer structure from tool names and descriptions alone, increasing hallucination risk and preventing confident downstream composition.
Tool annotations (readOnlyHint, idempotentHint) are absent despite all 16 tools being read-only. Without explicit readOnlyHint annotation, agents and clients cannot quickly determine which tools are safe to call without user confirmation or which can be batched.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | A | 83 | <=2025-11-25 | v2 |
Get App Store reviews. NUMERIC ID ONLY (no bundle id). Max page 10, 50 reviews/page (500 max per storefront). Empty result doesn't mean app missing. 1 credit/page.
Full Booking.com property: rooms, rates, facilities, policies. Send checkin+checkout together. 1 credit.
Booking.com guest reviews with score breakdown. No pagination. 1 credit.
Get full Capterra product profile with 25 most recent reviews included. Use get_capterra_reviews only to page past them. product_id (string, not number) or url required. 2 credits.
Page past the 25 reviews in get_capterra_product. 25/page, max page 100. slug is LOAD-BEARING here (case-sensitive; wrong slug silently returns page 1). product_id or url required. 2 credits/page.
Search Airbnb stays. ONLY source of prices (get_airbnb_listing has none). Send check_in+check_out together or prices default to random window. Prices are whole-stay totals, not per night. Paginate with page OR cursor, not both. 1 credit/page.
Search Amazon products. No sort option (always relevance). No category/price filter. 1 credit/page.
Search Apple App Store. NO PAGINATION: use limit (1-200) only. Searching a developer name returns their catalogue. 1 credit.
Search Booking.com stays. Requires destination or dest_id. checkin+checkout MUST be sent together. 25/page. 1 credit/page.
Search Capterra for B2B software. Max 20 results, no pagination. query or url required. 2 credits.
Parameter dependency documentation incomplete. Several tools have conditional parameter requirements (e.g., get_booking_hotel requires checkin+checkout sent together, get_capterra_reviews requires slug which is load-bearing) documented in descriptions, but not formalized in schema constraints or multi-parameter validation rules. This relies on LLM understanding rather than machine-enforced validation.
Error recovery guidance limited. The handleApiError() function returns generic error messages (e.g., 'Scavio API error (status): message') without suggesting next steps. For 404 errors, agents should be told to verify IDs; for 429 errors, to back off; for 402 (insufficient credits), to top up the account. Current messages are less actionable than they could be.
get_amazon_options description is minimal ('Free, no args.'). While accurate, it lacks context about when to call this tool or what structure it returns. A 70-character description does not help LLMs decide whether this is the right discovery tool.