Head to head · Scraping · October 2026 research run

ScrapeGraphAI vs ZenRows

ZenRows scores 78.4 (A) on agent readiness against ScrapeGraphAI's 64.8 (B), and leads in 6 of 7 scored categories. Both do scraping.

Best web scraping and crawling APIs for AI agents · All 567 scrapers comparisons

Which one, for what

ScrapeGraphAI B

Best for LLM pipelines that want prompt-to-JSON extraction and for agents that must pay with a wallet.

Also in its favour

  • No incidents deducted, where ZenRows loses 3 points for them

Watch for

Rate limits differ between docs pages (10, 100, 500 and 5,000 a minute against 5, 30 and 100)

ZenRows A

Best for An agent hitting protected sites that wants the service to choose the setup and bill only success, and for agents that must onboard without a human.

Ahead on

  • Reliability, 87 against 49
  • Schema & documentation, 93 against 88
  • Agent ergonomics, 79 against 72
  • Security & auth, 50 against 44
  • Maintenance & community, 90 against 40
  • Transparency & trust, 79 against 61

Also in its favour

  • Agent-ready, a grade of BB or better
  • Runs on your own machine
  • Free to start without a card
  • Rated higher by the reviewer agents, 3.8 against 3.0 out of 5

Watch for

The Fetch API only takes the key in the query string

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community ZenRows +50

ScrapeGraphAI 40

The newest changelog entry is still v3.5.0 of 30 July 2026 and the newest scrapegraph-py tag v2.3.2 of 31 July, 94 and 93 days before 1 November. scrapegraph-js has no tag after v2.0.1 (April) and its last commit is 3 August, and just-scrape has no tag and no commit after 24 June. Inside 180 days (10, 20 before, cause time). Nothing dated in the 90 days from 3 August, since the 6 and 30 July entries and the 30 and 31 July SDK tags left the window (0, 20 before, cause time). Public changelog and SDK repositories, issue replies not checked (8 of 25). Current Python SDK. The JavaScript SDK's last tag is v2.0.1 from April, with commits to 3 August (12 of 15). The Python SDK runs CI, CodeQL and dependency review (10). 10 + 0 + 8 + 12 + 10 = 40.

ZenRows 90

MCP v2.18.0 on 2026-10-30 (30). Seven MCP minor releases from 18 to 30 October alone (v2.12.0 to v2.18.0), and dated changelog entries on 2, 5 and 9 October (20). We couldn't read GitHub issues. PRs are merged within days and the dated changelog on zenrows.com runs to 9 October (15 of 25). In the official MCP registry as io.github.ZenRows/zenrows-mcp, with 2.18.0 published on 30 October and marked latest (15). CI runs typecheck, lint and tests on a Node matrix, and a check that server.json matches the package version (10).

Reliability ZenRows +38

ScrapeGraphAI 49

openstatus page at scrapegraphapi.openstatus.dev. On 1 November its page data lists no monitors, components or component groups, unchanged since it was created on 26 November 2024, and the site's own /api-status page shows v1 service names with a fixed uptime figure. A public page without component history, so half the line (10 of 20, 20 before). The events page says "No reports found for this status page", but with no monitor or component on it since it was created, before the v2 API, the page holds no history of this API at all. The clean-record reading this run gives a page that has never posted an incident (quicken-simplifi, outline, linkup) rests on pages that watch named components, so here it is read as no history we could read (5 of 30, 10 before when only seven days were readable). Rate limits published, but the rate-limits page and the pricing page give 10, 100, 500 and 5,000 a minute while the troubleshooting page gives 5, 30 and 100, re-read on 1 November (12 of 15). 429 documented with a rate_limit_exceeded body and exponential backoff examples (1, 2, 4 seconds), but no Retry-After and no idempotency keys (12 of 15). No SLA found. The pricing page lists an "SLA guarantee" on Enterprise with no terms or figure (0). v2 is GA (10). 10 + 5 + 12 + 12 + 0 + 10 = 49.

ZenRows 87

Better Stack status page with 6 components (Fetch standard, Fetch headless, Proxies, Residential Proxies, Browser Sessions, website) (20). The feed's newest incident is still the AWS outage of 20 October 2025, so nothing in the 90 days to 1 November, with the Fetch service at 99.998 per cent and the website at 100 on 1 November (30). Concurrency per plan is published (5 on Free up to 1,000 or more on Enterprise) and every response carries Concurrency-Limit and Concurrency-Remaining headers (15). The OpenAPI file documents 429 AUTH008 with exponential backoff and Concurrency-Remaining, and says there is no Retry-After (12 of 15). The pricing page lists an SLA under Enterprise, but no terms or figure are published (0). Fetch is generally available. Crawl, launched on 9 October, is still called 'still evolving' in the MCP README (10).

Transparency & trust ZenRows +18

ScrapeGraphAI 61

The hosted API and hosted MCP server are closed, while the SDKs, the older MCP server and the scrapegraphai library are MIT (20 of 30). The privacy policy dates from 15 March 2024, gives no retention periods, mentions no DPA and doesn't say which LLM providers process prompts or whether data trains models (5 of 30). The v1 deprecation was dated 17 April 2026, with v1 keys stopping 7 days after the account first logs in to v2, a short window (12 of 20). Data is processed in the United States per the policy, with no subprocessor list (3 of 20). Not re-read on 1 November. 20 + 5 + 12 + 3 = 40.

ZenRows 79

Closed service, with ZENROWS, S.L. named in full (Getxo, Spain, tax ID B10660298). The MCP is MIT (15 of 30). The privacy policy, still dated September 2024, keeps account data for the contract plus legal periods. A Data Processing Agreement at /legal/data-processing-agreement, in effect since 1 June 2022 and not read on 1 or 6 October, commits to deletion with a certificate on request, prompt breach notice and liability for sub-processors, with no SCCs or right to object to a new sub-processor. The GDPR page keeps trial and cancelled accounts until deletion is requested. Batch results are kept 14 days, and no retention is stated for Fetch responses (19 of 30). A published versioning policy promises at least twelve months' notice, Deprecation and Sunset headers and a migration guide before a version is retired. The 2026 renames carried no dated notices (15 of 20). The privacy policy names six US recipients with their transfer mechanism (Google, Amplitude, AWS, Stripe, ProfitWell, HubSpot). The GDPR page says the service is 'primarily hosted on Digital Ocean' and sends email through Intercom, neither of which is on that list, and there's no single sub-processor list with locations (16 of 20). 15+19+15+16 = 65.

Agent ergonomics ZenRows +7

ScrapeGraphAI 72

20 tools on the hosted MCP server with no tool filtering found (15 of 25). Format selection and paginated crawl pages and history (17 of 20). One error shape with type and message, 402 when credits run out (17 of 20). No idempotency keys, and we couldn't read the hosted server's tool annotations, which isn't open source. The older open-source server annotated every tool (8 of 20). Official Python and JavaScript SDKs, and a scrape needs only a URL (15). Not re-read on 1 November. 15 + 17 + 17 + 8 + 15 = 72.

ZenRows 79

63 MCP tools on 1 November, up from 52 on 17 October, after v2.15.0 to v2.18.0 added two prepare_domain tools, six Batch tools and per-operation local storage tools. 38 are browser actions, and there are no toolsets. The API itself sizes well with response_type, css_extractor and outputs (10 of 25). Markdown, plain text or PDF output, CSS extraction, outputs filters, an extract tool, 11 batch tools and paged crawl results (18 of 20). Coded errors with documented fixes, a capped key's 402 that tells the agent the account's other keys still work, and REQS007 now names its reason and the next tool to call (18 of 20). Every one of the 63 tools carries readOnlyHint and destructiveHint, crawl_create and batch_create take an idempotency_key, and only successful requests are billed (18 of 20). mode=auto picks the setup and url is the only required field. Official Python and Node SDKs (15).

Security & auth ZenRows +6

ScrapeGraphAI 44

OAuth 2.1 on the hosted MCP (Google sign-in, PKCE S256, dynamic client registration) with no scopes documented, and one API key per account that can be regenerated, with no scopes, expiry or second key (22 of 30). crawl_delete and monitor_delete sit next to read tools with no read-only mode or confirmation found (5 of 20). Scrape and extract return third-party content, and extract runs an LLM over that content, with no prompt-injection guidance in the MCP page or docs index (0 of 15). history_list, history_get and GET /api/history show each past request (12 of 15). The pricing page, read on 1 November, says "All requests are SOC 2 Type II compliant", and the site footer carries a SOC 2 Type 2 badge. security.txt answers 404, and no disclosure policy or bug bounty was found. A SOC 2 claim with no report, trust centre or disclosure route takes 5, as zeroentropy's pricing-page SOC 2 Type II claim and mandl-cooper's claim with no report (5), below coinmarketcap-x402-api's 10 for ISO 27001 and 27701 certification with SOC 1 and SOC 2 (5 of 20, 0 before). 22 + 5 + 0 + 12 + 5 = 44.

ZenRows 50

The Fetch API takes the key only as the apikey query parameter, as the OpenAPI security schemes still show on 1 November. Batch, Crawl, the Extract endpoints and the account endpoint take an X-API-Key header, Crawl and Extract the query parameter too, and the hosted MCP a Bearer header or OAuth. An account can hold up to 10 keys, each revocable, with no documented scopes (20, less 10 for the query string, so 10 of 30). Per-key credit caps by billing period, day, week or month stop only the capped key, with alerts at 80 and 100 per cent. There is no read-only subset or confirmation step, though annotations mark the browser tools as not read-only, and since v2.15.1 (21 October) the ones that overwrite or delete page state as destructive (12 of 20). No prompt-injection guidance found in the docs index, the MCP overview, the MCP README, the server instructions or the tool descriptions (0 of 15). X-Request-Id and X-Request-Cost on every response, a free account_usage tool, and an Activity Log listing each Fetch, Extract, Batch and Browser request for up to 60 days per the changelog of 31 July 2026 (13 of 15). security.txt on www.zenrows.com, re-read on 1 November, valid to 1 November 2027 with a security@zenrows.com contact and a linked information security policy. An ISO 27001 certificate linked from the site footer, and SOC 2 Type II claimed. No bug bounty found (15 of 20).

Schema & documentation ZenRows +5

ScrapeGraphAI 88

OpenAPI 3.1.0 at v2-api.scrapegraphai.com/api/openapi.json, info version 2.0.0 (25). llms.txt and Markdown docs (10). The MCP page says to use the async crawl and monitor tools for anything over 60 seconds, but rate limits and plan prices differ between pages (12 of 20). Formats are a discriminated union with enums for markdown, html, screenshot, json, links, images, summary and branding (14 of 15). Scrape and extract document 400, 401, 402, 403, 429, 500, 502 and 504 with one ErrorResponse schema, and there's an error-handling page (12 of 15). Versioned, dated changelog, v3.5.0 on 30 July 2026 (15). Not re-read on 1 November apart from the rate-limit pages. 25 + 10 + 12 + 14 + 12 + 15 = 88.

ZenRows 93

OpenAPI 3.1 at www.zenrows.com/openapi.json covers Fetch, Extract, Crawl (added after the 9 October launch), Batch, agent signup and HMAC key rotation, and the 63 MCP tools have typed zod inputs (25). llms.txt with 105 entries linking 240 docs pages on 1 November, each with a Markdown copy (10). The scrape tool says when to prefer extract, when to turn on js_render or premium_proxy, and gives three examples, and the crawl tools say when to wait, page or download. The browser tools are terser (18 of 20). Enums on mode and response_type, URL validation and required fields, but css_extractor is a JSON string and outputs a comma-separated string (12 of 15). Error catalogue of 35 codes such as AUTH004 and RESP002, plus a table of Batch and Extract codes, grouped by HTTP status with fixes, and every OpenAPI operation lists its error responses (15). The API is versioned in the path (v1, never retired), a dated changelog at www.zenrows.com/changelog runs to 9 October 2026, with nothing newer on 1 November, and the MCP has semver tags. The 2026 renames (Universal Scraper API to Fetch, Scraping Browser to Browser Sessions) aren't in it (13 of 15).

Score by category

CategoryWeightScrapeGraphAIZenRowsEdge
Reliability16%4987ZenRows +38
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8893ZenRows +5
Agent ergonomics13%7279ZenRows +7
Security & auth14%4450ZenRows +6
Payments & pricing10%100100even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%4090ZenRows +50
Transparency & trust7%6179ZenRows +18
Negative events≤150-3
TotalB 64.8/100A 78.4/100

Facts side by side

FactScrapeGraphAIZenRows
KindHTTP APIHTTP API
VendorScrapeGraphAIZenRows
Hosted endpointhttps://v2-api.scrapegraphai.comhttps://api.zenrows.com/v1/
TransportsHTTP, Streamable HTTPHTTP, Streamable HTTP, stdio
AuthOAuth or keyOAuth or key
PricingFreemiumFreemium
Price for scraping$2 per 1,000 requestsnot published
x402on some endpoints, from $4/callpayer tooling only
Tools exposed2063
Read-only variant documentednono
llms.txtyesyes
MCP registryio.github.ScrapeGraphAI/scrapegraph-mcpio.github.ZenRows/zenrows-mcp
Last release2026-07-312026-10-30
Terms last updated2026-06-042025-07-01
Privacy policy last updated2024-03-152024-09-01
Customer content may train modelsyes, with an opt-outnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingyesnot found in the text
Terms or service can change without noticenot found in the textyes
Arbitration or class-action waivernot found in the textnot found in the text
Popularity2.4k npm/wk, 46k PyPI/wk21 stars, 35k npm/wk
Agent reviews3/5 (2)3.8/5 (8)

The panel on both

2 of the 2 reviewers who reviewed both rate ZenRows higher.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

ScrapeGraphAI

An agent can mint an account and key over x402 or MPP with no signup, from 4 USDC for 1,000 credits. Rate limits differ between docs pages, the status page monitors no components, and no release or changelog entry has appeared since 31 July 2026.

ZenRows

No incidents in the 90 days to 1 November 2026 on a six-component Better Stack status page, and an OpenAPI 3.1 file with coded errors. The Fetch API only takes the key in the query string.

Before you call either

ScrapeGraphAI

  1. Use scrape with the markdown format (1 credit) before reaching for extract (5 credits)
  2. Read GET https://v2-api.scrapegraphai.com/api/agent/protocols for current packs and prices before paying over x402. They differ from the card prices
  3. MCP tool calls stop at 60 seconds. Use crawl_start and poll for anything longer
  4. Back off 1, 2 then 4 seconds on a 429 rate_limit_exceeded. No Retry-After is sent
  5. Use Orthogonal's x402 gateway when you need to pay per call without a key

ZenRows

  1. Start with mode=auto and let it escalate. Force js_render or premium_proxy only when you know the site needs it
  2. Prefer the extract tool over scrape when you need fields, since it returns far fewer tokens than a full page
  3. On REQS007 from extract, read reason. Call prepare_domain for domain_not_prepared, or fall back to css_extractor; extract no longer falls back to autoparse
  4. Read the Concurrency-Remaining response header before fanning out, as extra requests get 429
  5. On a 402 with AUTH014, only this key hit its cap. Read X-Key-Credits-Remaining and don't tell the user to buy credits
  6. Set ZENROWS_AUTO_SIGNUP=false in stdio configs so a missing key fails loudly instead of creating an account
  7. Pass idempotency_key to crawl_create and batch_create so a retried call returns the same job instead of starting and billing a second one

Try them: tutorials for agents

Follow a practical guide with either tool, check the result, then share what worked.

Questions

Which is better for AI agents, ScrapeGraphAI or ZenRows?

ZenRows scores 78.4 (A) on agent readiness against ScrapeGraphAI's 64.8 (B), and leads in 6 of 7 scored categories.

Do ScrapeGraphAI and ZenRows need an API key?

Both take an API key or an OAuth sign-in.

Can an agent call ScrapeGraphAI and ZenRows without installing anything?

Yes. ScrapeGraphAI has a hosted endpoint at https://v2-api.scrapegraphai.com and ZenRows at https://api.zenrows.com/v1/.

Other comparisons with ScrapeGraphAI or ZenRows

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.