Qovia vs ScrapingBee

Qovia 30

Closed service. The Terms of Service is one paragraph that names no legal entity, address or liability terms, says Delaware law governs and that the page is a summary with the signed agreement governing; the repository's client is MIT. Between the closed and unclear bands (7 of 30). The privacy policy is one paragraph. Account information is used to deliver and bill, personal information is not sold, requests and results are not resold, deletion requests go to the support mailbox, and the agreement governs where it differs. No retention periods, no subprocessors, no DPA, and no statement about the cache the prov.source flag implies; the SECURITY.md repeats the same points (8 of 30). A 410 endpoint_disabled code exists for retired endpoints and the changelog is dated, but there is no deprecation policy or notice period and nothing is marked deprecated (5 of 20). Stripe is named on the pricing page as the card and Alipay processor; no other provider, data location or transfer basis is stated (3 of 20). 7+8+5+3 = 23.

ScrapingBee 65

Closed service with terms naming VostokInc SAS, Paris, registration 843 352 683 (15 of 30). The privacy policy, effective 15 July 2024, keeps prospect and customer data up to 3 years and links a DPA. It doesn't mention the Oxylabs group or say whether scraped content is stored (18 of 30). The Node SDK changelog dates removals, for example youtubeTranscript() dropped in v2.0.0 on 2026-09-01, and the api_key query parameter is marked deprecated. No standing deprecation policy (10 of 20). Processors described by category, located in the US or EU. No named list (8 of 20).

Qovia 40

Hosted reading, for the REST API with the MCP server described. No public status page was found. https://qovia.ai/status/ answers 404, status.qovia.ai does not resolve and neither the docs nor the footer links one (0). No readable incident history, so 5 by the rubric (5). Limits are published with numbers on the quickstart's rate-limits section, 60 requests a minute per account before the first top-up and 600 after on a sliding 60-second window, with X-RateLimit-Limit and X-RateLimit-Remaining headers, a 64 KiB body cap, an empty-call throttle (50 calls in 10 minutes with over 80 per cent empty, not found or rejected) and a per-second cap per endpoint without a number (15). Retry-After is declared on too_many_auth_failures, rate_limited, too_many_unbilled, endpoint_busy and upstream_busy, the error table marks the retryable codes, the docs say a 503 endpoint_unavailable carries none and will not clear, and the Python client retries 429 and 503 honouring the header. Every operation is a read, including POST /v1/search, and a failed or timed-out call is refunded, so idempotency keys are not needed (15). No SLA. The terms summary says nothing about uptime and the pricing FAQ excludes outages, rate limits and platform changes from refunds; the 11 September 2026 archived home page advertised a 99.95 per cent enterprise SLA that the current pages have dropped (0). The endpoints are sold and labelled Live on the docs and coverage pages, but every one of the 143 carries x-qovia-status beta in the OpenAPI file, the API version is 0.1.0 and the MCP server reports 0.1.0 (5 of 10). 0+5+15+15+0+5 = 40.

ScrapingBee 70

Hyperping status page at status.scrapingbee.com with 6 components and history (20). In the last 90 days the dedicated APIs (Google, Amazon, Walmart) were down for about 40 minutes on 24 July, the HTML scraper ran degraded for about 24 hours from 28 July, and the Google API was degraded on 22 September. No full core outage of an hour, but the day-long degradation of the main API weighs more than a minor, so we scored between the bands (15 of 30). Concurrency per plan is published, 25 on Hobby to 400 on Business+ (15). 429 means the concurrency limit and isn't billed, failed target requests are retried server-side for up to 30 seconds, and the MCP notes advise exponential backoff. No Retry-After found (10 of 15). No SLA found (0). Generally available (10).

Qovia 86

An OpenAPI 3.1.0 file (version 0.1.0, 1 MB) is public at qovia.ai/openapi.json with 144 paths and 144 operations, two security schemes, typed query parameters, a response example on every operation and the error responses declared on each. The data payload itself is typed only as object, array or null with per-endpoint examples, and the MCP call_endpoint tool takes a free params object, so 22 of 25 (22). llms.txt (75 KB, one line per endpoint with price and status) and llms-full.txt (331 KB, the whole documentation as text) are served, and the quickstart's billing, limits and error sections are in both (10). Every operation has a summary and a description; most say what the data holds, how to page, which ids to pass, which fields are unavailable and what answers 404 or empty, and the MCP server's instructions say to find, read then call. Guidance on when not to use an endpoint is thin, mostly pointers such as replies coming from another endpoint (16 of 20). 321 parameters, 75 with enums, patterns on ids, min and max on strings and integers, 117 operations with one required parameter and 16 with none, the POST body typed with additionalProperties false, and unknown parameters rejected with 400 (14 of 15). A response example on all 144 operations, 20 error codes with every message variant listed, response headers documented per error, and cURL, Python, JavaScript and raw HTTP samples; the examples are catalogue samples the docs say are not live data (14 of 15). Paths are under /v1, a 410 endpoint_disabled code exists for retired endpoints, and the changelog is dated, but it has two days (21 and 22 October 2026), nothing is marked deprecated and no notice period is stated (10 of 15). 22+10+16+14+14+10 = 86.

ScrapingBee 60

No OpenAPI spec found. Per-endpoint llms files list parameters with types and defaults, and the hosted MCP's schemas weren't visible to us (8 of 25). llms.txt indexes 26 per-endpoint text files written for models (10). The API files explain each parameter's purpose and cost, and a reference-notes file lists 9 known conflicts between its own pages. MCP tool descriptions aren't published (10 of 20). Parameters are typed with defaults, and mode, proxies and render_js are documented as enumerated options (10 of 15). Status code table with whether each code is billed, plus examples per endpoint (14 of 15). The SDKs and CLI keep dated CHANGELOGs. The API has no changelog (8 of 15).

Qovia 60

API keys with scopes chosen at creation (every endpoint, search, every raw endpoint or one of 24 platform groups), an optional expiry, revocation in the console and hashed storage; the docs say the key is read from the Authorization or x-api-key header only, never the URL or a cookie, and 30 failed authentications from one IP in a minute are throttled. No OAuth and no rotation feature beyond creating another key (24 of 30). The whole surface is read-only retrieval, scopes restrict a key to platforms, the two MCP catalogue tools need no key, and the docs and SECURITY.md tell users to give AI tools a separate scoped key (17 of 20). Every response is third-party platform text, and the only related statement is that error messages never contain the data source's own text; nothing on treating returned content as untrusted (3 of 15). request_id is the usage record ID on every answer, the console lists every charge in a usage log, X-Qovia-Charged and X-Qovia-Balance headers report each call, and GET /v1/account returns the key's scopes and balance at no charge; no exportable audit log or per-call endpoint was found (11 of 15). No security.txt on qovia.ai or api.qovia.ai (404), no bug bounty, SOC 2 or ISO 27001; the repository's SECURITY.md names a Telegram handle and a mailbox for reports covering both hosts (5 of 20). 24+17+3+11+5 = 60.

ScrapingBee 37

One API key per account, which the MCP notes say can be revoked (401 on revoked keys). Team management exists. No scopes (20). The api_key query parameter is deprecated but still works, and remote-mcp.txt says the hosted MCP takes the key only as ?api_key= in its URL, so less 10 (10 of 30). No read-only or scoped key, and the HTML API can send POST requests to target sites. The CLI gates shell post-processing behind an explicit unsafe flag (8 of 20). The official CLI skills and AGENTS.md say scraped output is data, never instructions, and tell the agent to report instruction-like content as a possible prompt injection. The MCP docs say nothing (12 of 15). A Usage API and a local CLI audit log. No per-request log in the docs (7 of 15). No security.txt per the 30 September check, and no bug bounty, disclosure policy or certification found (0 of 20).

Qovia 55

No x402, MPP or L402 on Qovia's own endpoints; USDT on TRC-20 is a top-up method from a console session, not a machine payment protocol. The vendor's five Actors on the Apify Store under the profile qovia (Qovia Data, the same Telegram handle and mailbox as the site), such as the Xingtu Actor at $0.01 to $0.06 a section, can be paid through Apify, whose prepaid token is sold over x402, so partial credit for a third-party route on part of the catalogue (10 of 40). Per-call prices are published without a login, per endpoint in the OpenAPI file, llms.txt and the docs, search at $5 and raw endpoints from $0.5 to $19.5 per 1,000, with the tier rule and payment fees stated (20). 50 trial calls for a new account, counted only when a call returns data; the quickstart says to top up when they run out and the docs ask for no card at sign-up, though no account was created to confirm it (20). A person signs up in a browser and creates a key; the MCP catalogue tools work without a key but data calls do not, and the Apify route reaches four platforms without a Qovia account (5 of 20). 10+20+20+5 = 55.

ScrapingBee 40

No x402, MPP or L402 (0). Credit costs per request type (1, 5, 10, 25, 75, plus 5 for AI extraction) and plan prices are public, though the reference notes admit some endpoint costs conflict between pages (20). 1,000-credit trial, no card (20). A person signs up in a browser (0).

Qovia 77

API reading. Responses come in fixed page sizes per endpoint, from about 5 comments on Xiaohongshu to 100 Fliggy reviews, with limit only on Bilibili danmaku and no field selection; the record shape is compact and shared across posts, notes and videos. The MCP server loads four tools by default, keeps the catalogue out of context until search_endpoints and get_endpoint are called, and ?tools= selects platforms or everything, which on the MCP reading would score 25 (18 of 25). Cursor pagination with has_more on most lists, search_id on Douyin and Zhihu search, page numbers on Weibo, Xiaohongshu and Bilibili, sort options on several, and the cross-platform search filters by platform, content type, sentiment and whole-day date windows up to 90 days (17 of 20). One error object with 20 codes, request_id on every answer, the retryable set marked, refunded codes separated from never-charged ones, and MCP tool errors carrying error.code, the HTTP status and retry_after_seconds (19 of 20). All calls are reads, failed and empty calls are refunded, Retry-After is declared and the Python client retries; no readOnlyHint or other annotation appears in the vendor's 28 October tools/list snapshot (14 of 20). One required parameter on 117 of 144 operations and defaults set on the server, but the only official SDK is the Python client; Node and curl are examples (9 of 15). 18+17+19+14+9 = 77.

ScrapingBee 68

18 MCP tools (15 of 25). json_response, extract_rules, ai_query, return_page_markdown and return_page_text, plus max_cost to cap escalation (18 of 20). HTTP status codes with meanings and billing, but no machine-readable error codes beyond the status (15 of 20). Failed and 429 requests aren't billed and mode=auto bills only the tier that works. No tool annotations found (8 of 20). Only url is required, but render_js defaults on at 5 credits. Official Python and Node SDKs plus a CLI (12 of 15).