Crawl4AI vs SoldFetch

Crawl4AI 83

Library v0.9.5 on 14 October 2026, 23 days before 6 November, and the cloud added /serp on 8 October (30). v0.9.3 (31 August), v0.9.4 (23 September) and v0.9.5 (14 October) in the 90 days from 8 August (20). 56 open issues on 6 November, the newest triaged with Bug, Needs Triage, In-progress, Root caused or Done labels within days, plus a Discord. Reply authors weren't read (20 of 25). Not in the official MCP registry, and the cloud has no official SDK; the Python library is current on PyPI (8 of 15). The security workflow runs a test suite and a default-posture gate, and releases are built and checked with twine. No general test workflow on push found (5 of 10). 30 + 20 + 20 + 8 + 5 = 83.

SoldFetch 49

Closed service. The latest dated changes are commits to all three public repositories and four blog posts on 27 October 2026, and the docs sitemap dates the authentication page 30 October 2026 (30). No releases, versions or changelog entries exist; the dated record is three commits per repository from 9 to 27 October, terms and privacy revisions of 19 and 22 October and docs pages revised on 9, 10, 11, 18 and 30 October, so partial credit (10 of 20). No public changelog; support is a contact form and contact@soldfetch.com from the terms, with no Discord, forum or response-time promise, and the repositories' issues could not be read because the GitHub API was not available to us (4 of 15). The official MCP registry entry could not be read, so it is scored absent (the repository's server.json names io.github.ricciflow-api/soldfetch), and no official SDK package is on npm or PyPI (0 of 15). Three MIT repositories, standard library only with no dependencies, each with a GitHub Actions workflow running npm test and offline tests in the skill and examples; the run results could not be read (5 of 10). 30+10+4+0+5 = 49.

Crawl4AI 70

The library is Apache-2.0 and the cloud is closed under terms that name CONTEXT4AI PTE LTD (22 of 30). The privacy policy gives retention periods (account plus 90 days, usage records 13 months, logs 90 days), but none for crawled content, and llms.txt says repeated URLs are served from a shared archive while the security page says fetched pages are used only to answer the request (18 of 30). The legacy cloud API closed on 6 November 2026 with a migration guide but no dated notice found, and the terms promise notice of material changes only 'where reasonable' (8 of 20). Subprocessors named with their locations on the security page, with EU storage of account records (18 of 20). 22 + 18 + 8 + 18 = 66.

SoldFetch 41

Closed hosted service. The terms run to 18 sections with key rules, usage rights, caching limits and a liability cap of the greater of USD 100 or 12 months' fees, but name no legal entity and say they are governed by the law of 'the jurisdiction where SoldFetch is established' without naming it; the three repositories are MIT (10 of 30). The privacy policy (updated 22 October 2026) gives request-log retention of 7, 30 or 90 days by plan, about an hour for exports, digests for keys and cookie durations, and the docs give cache periods and 24 hours for job results; the pricing page's 30-day request logs on every plan disagrees with the policy's plan-based periods, backups have no period and no DPA was found (20 of 30). No deprecation policy or dated notices; terms 11.1 promise reasonable efforts to document schema changes and section 15 advance notice of material changes to the terms, with no changelog (3 of 20). Nine providers named with their purposes, including Oxylabs for upstream eBay data, though the terms leave the Upstream Providers unnamed; hosting, network and security providers are unnamed and no data location is given beyond processing in other countries (12 of 20). 10+20+3+12 = 45.

Crawl4AI 35

Graded on the hosted lines, for Crawl4AI Cloud, which the README and llms.txt now lead with and which an agent calls through its MCP server; the library would read on the local lines. No status page is linked, and status.crawl4ai.com fails TLS (0). No incident history readable (5). Per-plan limits are published with numbers in /v1/prices, from 10 requests a minute and 2 at a time on Free to 1,500 and 50 on Scale (15). 429 carries X-RateLimit headers and a message, 503 busy says to resend after a few seconds, and MCP job submissions take an idempotency key, but no Retry-After or backoff guidance (10 of 15). No SLA found in the terms or pricing (0). Prices and the free credit are a soft launch, /serp is beta and /answer experimental, while scrape, extract and search carry no beta label (5 of 10). 0 + 5 + 15 + 10 + 0 + 5 = 35.

SoldFetch 55

Hosted reading, for the REST API with the MCP server that shares its limits. SoldFetch's own status page at https://soldfetch.com/status lists ebay.com, ebay.co.uk and ebay.de as components, checked every five minutes, all operational on 3 November 2026, but keeps no incident log or uptime history, so half credit (10 of 20). No readable incident history, so the floor (5). Limits are published with numbers, 10 data requests a minute and 1 concurrent on Basic and 60 a minute and 5 concurrent on paid plans, per workspace across keys and interfaces, with a separate 60 a minute for usage and history (15). 429 is documented for workspace rate, concurrency and quota limits, 503 upstream_rate_limited carries a Retry-After header in seconds and the error schema an optional retryAfter, the docs ask for bounded backoff with jitter and name what not to retry, and REST data calls take an Idempotency-Key with a free replay for 24 hours and 409 on changed inputs; job submissions have no key, but the docs say to check existing jobs before resubmitting and allow one active job per workspace (15). No SLA on self-service plans; terms 11.2 keep uptime commitments to a signed enterprise order form (0). No beta or preview label on the API or the MCP server; three of the eight recognised eBay sites are enabled 'at launch' (10). 10+5+15+15+0+10 = 55.

Crawl4AI 46

Bearer keys, shown once and regenerable, with 3 to 30 keys per plan. claude.ai and ChatGPT sign in through a flow that advertises one scope, mcp, and each app shows as a revocable key. No per-key scopes, and the docs don't offer the key in a URL (22 of 30). A key can call billing routes, and the MCP topup and recharge tools charge the saved card, though llms.txt tells agents to ask first and accounts can set a spend cap. No read-only key (6 of 20). Scraped pages are untrusted text and no prompt-injection guidance was found in llms.txt or the API docs (0). Per-call cost, balance and fetch path in response headers and fields, no access log documented (6 of 15). The security page (4 October 2026) gives a disclosure process with fix targets and says SOC 2 is in progress, and the library publishes GitHub advisories with reporter credits. security.txt redirects to a 404, and no bounty found (12 of 20). 22 + 6 + 0 + 6 + 12 = 46.

SoldFetch 62

Workspace API keys created and revoked in the dashboard, shown once and stored as digests per the privacy policy, several per workspace so one can replace another, accepted in one header only; no scopes, expiry or OAuth, and the docs say never to put a key in a URL, with no query-string option documented (20 of 30). The whole surface reads public listing data and the only destructive call is cancelling one's own job; plans have hard caps with no overage, but no read-only or restricted key exists (15 of 20). Every response carries third-party listing text. The AI agents guide, the MCP README, SECURITY.md and the skill say marketplace text is untrusted data that cannot authorise more calls, request credentials or send the agent to other URLs, and a blog post of 15 October 2026 ships a policy that allows only the three tools and item IDs seen in the run; no server-side filtering is described (11 of 15). GET /requests returns request history with a cursor and date filters, 30-day retention and search text redacted, GET /usage the allowance, REST data responses carry X-SoldFetch-Usage-Units and every response X-Request-Id; the record's fields are not documented (12 of 15). No security.txt (client error), bug bounty, SOC 2 or ISO 27001; the repositories' SECURITY.md files give a reporting route to contact@soldfetch.com (4 of 20). 20+15+11+12+4 = 62.

Crawl4AI 55

No OpenAPI or similar contract for the cloud API, and the MCP tool schemas are only in the live tools/list, which needs a key. /v1/prices is machine-readable but covers prices, not calls (0). llms.txt of 304 lines written for agents (10). Each endpoint says what it is for and when to choose another, such as trying without a proxy first, using browser only when auto misses content and /scrape/jobs for slow lists (16 of 20). Parameters are listed with types, defaults, allowed values (format both, md or html; proxy none, isp or residential) and limits, in prose (10 of 15). Curl examples per endpoint and documented error strings per status with whether each is charged (14 of 15). No changelog for the cloud API, only dated notes such as /serp 'added 8 Oct 2026'. The library has a dated CHANGELOG (5 of 15). 0 + 10 + 16 + 10 + 14 + 5 = 55.

SoldFetch 68

API reading, with the MCP server described. An OpenAPI 3.1 file is published at https://soldfetch.com/api/v1/openapi.json, under a path robots.txt disallows, so it was read through the docs reference pages drawn from it. Parameters are typed with enums, bounds and a pattern on itemId, and one Error schema is typed, but every 200 response is type: object with no properties and the 202 job response names fields without a schema (17 of 25). llms.txt on the site and the docs, llms-full.txt, and Markdown twins of the docs pages (10). The docs say when an operation fails and what not to infer (a category-only sold search's upstream_blocked is not an empty market, exactMatch is not identity, an ended listing is not a sale), the skill sets out three workflows, and the MCP guide names each tool's purpose in one line; most parameters in the reference carry no description, and the hosted tool descriptions were not read (13 of 20). ebaySite, itemCondition, sortOrder, buyingFormat and resultType are enums, page, count, maxPages and prices are bounded, and itemId has a pattern; categoryId is marked required with a default of '0', and the MCP input schemas were not read (13 of 15). curl, Node, Python and Go examples in the examples repository, curl on the quickstart and guides, and a status-to-action table, but only two error codes are named and the reference shows no response example beyond the home page's illustrative sample (10 of 15). Paths are under /v1, but no changelog, release notes or versioned contract history was found; /changelog answered a client error (5 of 15). 17+10+13+13+10+5 = 68.

Crawl4AI 67

16 MCP tools, six of them for billing and support (15 of 25). format md or html, parse opt-in blocks, batch limits and job results paged 500 at a time over HTTP and 50 over MCP (16 of 20). String error codes an agent can branch on (blocked, login-wall, no-answer, busy, no_credit, plan_cap, spend_cap), each with whether it is charged (18 of 20). Failed and cancelled calls aren't charged and jobs_submit takes an idempotency key, but no tool annotations are documented and two tools can charge the saved card (10 of 20). Only url or q is required. There is no official SDK for the cloud API; the docs give curl and standard-library Python and Node examples (8 of 15). 15 + 16 + 18 + 10 + 8 = 67.

SoldFetch 75

API reading. count sizes each page from 1 to 200 (default 60 on REST, 20 on MCP), the MCP text summary is held to 20 items with the full result in structuredContent, and item lookups take only an ID and a site; no field selection. On the MCP reading three tools would score 25 (20 of 25). Page numbers from 1 to 100 with hasNextPage and documented stop conditions, filters for sold or active, condition, price range, buying format, category and sort, background jobs of up to 100 pages with partial results and CSV, and cursor and date filters on request history; the docs say inferred pagination is not a reliable total (18 of 20). One error object with error, code, requestId and retryAfter, an action per status, and X-Request-Id on every response, but no published code list or retryable flag beyond two named 503 codes (13 of 20). Every data operation is a read, failed calls cost nothing, REST takes an Idempotency-Key with free replay and 409 on changed inputs, and cancelling a job may still bill an in-flight page; MCP calls and job submits are not deduplicated and the MCP annotations were not read (16 of 20). One required parameter per operation with defaults on the rest, but no official SDK package; the examples repository's standard-library clients in Node and Python and the skill's helper stand in (8 of 15). 20+18+13+16+8 = 75.