# Auxiliar.ai

> One API key for 24 search, scraping, browser-automation, and voice APIs — unified billing, no subscriptions, instant access. For you and your AI agents.

## About
Auxiliar.ai is a web-access API gateway for AI agents. Reach 24 search,
scraping, browser-automation, and voice providers through one API key and one
billing account — no per-provider signups or subscriptions. Upstream provider
keys are injected server-side, so your agent never handles them; usage is
credit-metered against your auxiliar account at each provider's real price.

## Calling the gateway
- Base URL: https://api.auxiliar.ai — authenticate with `Authorization: Bearer <your key>`.
- Route rule: `/{provider}/{provider-native-path}`. The gateway forwards to the
  provider's native API and returns its response unmodified — same paths, same
  parameters, same response shape as the provider's own documentation.
- Example (Serper search):
  curl https://api.auxiliar.ai/serper/search \
    -H "Authorization: Bearer $AUXILIAR_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"q": "your query"}'
- Failure is deterministic: an empty balance returns HTTP 402 until credits are
  added — never a silently degraded response. Optional auto-top-up keeps
  unattended agents funded; every request is debited individually in the ledger.
- Sign-up: https://accounts.auxiliar.ai/sign-up — $5 free credit, no card, and
  your first API key is minted automatically. Keys, ledger and auto-top-up live
  at https://panel.auxiliar.ai
- The HTTP gateway is the production surface — any provider's existing SDK works
  by pointing it at the base URL. Optional MCP discovery shim: `npx auxiliar-mcp`.

## Why route through the gateway — and when not to
- Search cost: host-model native web search typically bills ~$10 per 1,000
  searches ($0.01/search) plus result-token overhead. Serper through the
  gateway is $0.0003/search — roughly 30x less. Reproduce it: the curl above
  is one $0.0003 call against your $5 free credit.
- Anti-bot scraping: on the shared anti-bot corpus (measured 2026-06-20),
  Firecrawl, Scrapfly and Zyte returned clean markdown on 100% of pages that
  block a plain fetch. Per-provider bypass rates: /tools pages and evals.json.
- Capabilities you likely don't have natively: text-to-speech (/elevenlabs)
  and driveable headless browser sessions (/browserbase) on the same key.
- No lock-in: passthrough means provider-native paths, params and response
  shapes — no proprietary SDK, no translation layer. To leave, point your
  code back at the provider's own base URL. Zero rewrite.
- No single-provider dependence: every verb has multiple ranked providers
  behind the same key. If one degrades, switch by changing one path
  segment — no new account, no new key.
- Interrogate before you sign up: the benchmark data is public
  (https://auxiliar.ai/evals.json) and the MCP advisor tools
  (recommend_tools, get_scorecard) need no API key.
- When NOT to use us: skip the gateway for providers you already hold a
  direct account with negotiated volume pricing — there the 5.5% top-up fee
  buys you nothing.

## Provider catalog — routes, prices, measured evals
Every provider on the gateway, with the measured numbers behind the
/tools rankings. All providers run the identical task corpus per verb;
run dates are per scorecard. Latest run: 2026-06-22.

## Search

### Brave Search — `/brave`
Independent crawler index · Search · tier T1
A genuinely independent 30B+ page index — web, image, video and news verticals.
Entry price: $5 / 1k · MCP: official
Choose if: You want an index that isn't a Google/Bing scrape, plus a privacy posture and AI grounding.
Avoid if: You need more than ~200 results per query.
Call: https://api.auxiliar.ai/brave/<native-path> — same paths, params and responses as Brave Search's own API.
Docs: https://docs.auxiliar.ai/brave · Eval page: https://auxiliar.ai/tools/brave/
Measured evals (curator fleet, shared public corpora):
- Answer — composite 7.02/10, rank #5 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.28 · Latency p50 2.4 s · Cost $0.06 / call · Errors 0% — Correct, but far pricier per query than peers.
- Search — composite 6.99/10, rank #4 of 11 (measured 2026-06-20): Recall@10 0.64 · Latency p50 1.3 s · Cost $0.0078 / useful result · Errors 0%

### DataForSEO — `/dataforseo`
SEO & SERP data platform · Search · tier T2
SEO and SERP intelligence — rankings, keywords, backlinks — with the richest SERP-element taxonomy.
Entry price: $0.60 async / $2 live / 1k · Latency: ~5s std / ~6s live · MCP: official
Choose if: You need bulk async SERP scraping, webhooks, and the deepest SERP-feature taxonomy.
Avoid if: You want low latency or a shallow response — deep tasks[].result[].items[] nesting is real friction.
Call: https://api.auxiliar.ai/dataforseo/<native-path> — same paths, params and responses as DataForSEO's own API.
Docs: https://docs.auxiliar.ai/dataforseo · Eval page: https://auxiliar.ai/tools/dataforseo/
Measured evals (curator fleet, shared public corpora):
- SERP verticals — composite 8.09/10, rank #4 of 4 (measured 2026-06-22): Quality 0.93 · Latency p50 8.2 s · Cost $0.002 / call · Errors 7%
- Search — composite 5.86/10, rank #10 of 11 (measured 2026-06-20): Recall@10 0.59 · Latency p50 2.9 s · Cost $0.0034 / useful result · Errors 0%

### Exa — `/exa`
Neural + keyword index · Search · tier T1
Neural/semantic search by meaning, with find-similar, a built-in answer endpoint and clean extraction.
Entry price: $7 / 1k (+content $1/1k) · MCP: official
Choose if: You want semantic discovery, 'find similar', research, or a built-in /answer endpoint.
Avoid if: You need full keyword/SERP parity — content and summary bill per page per type and stack.
Call: https://api.auxiliar.ai/exa/<native-path> — same paths, params and responses as Exa's own API.
Docs: https://docs.auxiliar.ai/exa · Eval page: https://auxiliar.ai/tools/exa/
Measured evals (curator fleet, shared public corpora):
- Answer — composite 7.9/10, rank #1 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.48 · Latency p50 1.5 s · Cost $0.005 / call · Errors 0% — Best citation faithfulness.
- Search — composite 6.7/10, rank #6 of 11 (measured 2026-06-20): Recall@10 0.61 · Latency p50 1.6 s · Cost $0.014 / useful result · Errors 0%

### Jina — `/jina`
Search + read-to-markdown · Search · tier T2
Best-in-class URL→clean-markdown Reader, plus a search endpoint that returns full-page markdown in one call.
Entry price: ~token-based · Latency: low RPM · MCP: official
Choose if: You want clean LLM-ready markdown from a URL or a search-plus-read in a single call.
Avoid if: You need high search RPM or many results per query (≤5 default, low RPM).
Call: https://api.auxiliar.ai/jina/<native-path> — same paths, params and responses as Jina's own API.
Docs: https://docs.auxiliar.ai/jina · Eval page: https://auxiliar.ai/tools/jina/
Measured evals (curator fleet, shared public corpora):
- Parse · PDF/doc — composite 8.43/10, rank #2 of 3 (measured 2026-06-22): Text accuracy 0.78 · Latency p50 263 ms · Cost $0 / useful · Errors 0% — Returns empty on scanned PDFs (no OCR).
- Search — composite 7.69/10, rank #1 of 11 (measured 2026-06-20): Recall@10 0.60 · Latency p50 1.4 s · Cost $0.00033 / useful result · Errors 0% — Wins all three query-classes.
- Answer — composite 6.61/10, rank #7 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.11 · Latency p50 2.6 s · Cost $0.02 / call · Errors 0%

### Linkup — `/linkup`
Own crawl index · Search · tier T2
A ToS-clean search alternative with native sourced answers, structured JSON and a /fetch markdown endpoint.
Entry price: $5 / 1k · MCP: official
Choose if: You want a compliant SERP alternative with a built-in sourced answer and structured JSON.
Avoid if: You need a relevance score, per-result date or geo controls (none exposed).
Call: https://api.auxiliar.ai/linkup/<native-path> — same paths, params and responses as Linkup's own API.
Docs: https://docs.auxiliar.ai/linkup · Eval page: https://auxiliar.ai/tools/linkup/
Measured evals (curator fleet, shared public corpora):
- Answer — composite 6.91/10, rank #6 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.16 · Latency p50 2.5 s · Cost $0.005 / call · Errors 0%
- Search — composite 5.47/10, rank #11 of 11 (measured 2026-06-20): Recall@10 0.39 · Latency p50 3.2 s · Cost $0.013 / useful result · Errors 0%

### Parallel — `/parallel`
Agent-optimized search index · Search · tier T2
An agent-optimized search index where a natural-language objective shapes ranking; returns pre-compressed excerpts.
Entry price: $5 (base) / $9 (pro) / 1k · Latency: P50 <1s · MCP: official + no-key MCP
Choose if: You want the lowest-friction in-loop option — a no-key Search MCP and sub-second P50.
Avoid if: You need a relevance score, full content or a synthesized answer in the Search response.
Call: https://api.auxiliar.ai/parallel/<native-path> — same paths, params and responses as Parallel's own API.
Docs: https://docs.auxiliar.ai/parallel · Eval page: https://auxiliar.ai/tools/parallel/
Measured evals (curator fleet, shared public corpora):
- Answer — composite 7.58/10, rank #3 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.48 · Latency p50 16.3 s · Cost $0.01 / call · Errors 0%
- Search — composite 6.76/10, rank #5 of 11 (measured 2026-06-20): Recall@10 0.64 · Latency p50 2.7 s · Cost $0.0062 / useful result · Errors 0%

### SearchAPI.io — `/searchapi`
Multi-engine SERP scrape · Search · tier T2
Multi-engine search results — Google, Bing, YouTube and 80+ engines — via one endpoint.
Entry price: $1.00–$4.00 / 1k · Latency: real-time · MCP: official
Choose if: You want SerpApi-compatible JSON at a lower price, with pay-per-success billing.
Avoid if: You depend on large Google result pages (num silently capped at 10 since Sept-2025).
Call: https://api.auxiliar.ai/searchapi/<native-path> — same paths, params and responses as SearchAPI.io's own API.
Docs: https://docs.auxiliar.ai/searchapi · Eval page: https://auxiliar.ai/tools/searchapi/
Measured evals (curator fleet, shared public corpora):
- SERP verticals — composite 9.02/10, rank #2 of 4 (measured 2026-06-22): Quality 0.98 · Latency p50 1.9 s · Cost $0.004 / call · Errors 0% — Broadest vertical coverage (SerpApi-compatible JSON).
- Search — composite 6.44/10, rank #8 of 11 (measured 2026-06-20): Recall@10 0.57 · Latency p50 2.5 s · Cost $0.007 / useful result · Errors 0%

### SerpApi — `/serpapi`
Multi-engine SERP scrape · Search · tier T1
Structured results from Google and 80+ other search engines, with the key handled for you.
Entry price: ~$15 → $9 / 1k · Latency: ~1.7–2.0s · MCP: official
Choose if: You're compliance-sensitive, need multi-engine breadth, or want the drop-in LangChain default.
Avoid if: You're price-sensitive versus Serper for Google-only queries.
Call: https://api.auxiliar.ai/serpapi/<native-path> — same paths, params and responses as SerpApi's own API.
Docs: https://docs.auxiliar.ai/serpapi · Eval page: https://auxiliar.ai/tools/serpapi/
Measured evals (curator fleet, shared public corpora):
- SERP verticals — composite 9.0/10, rank #3 of 4 (measured 2026-06-22): Quality 1.00 · Latency p50 1.4 s · Cost $0.015 / call · Errors 0% — Covers every vertical tested.
- Search — composite 6.1/10, rank #9 of 11 (measured 2026-06-20): Recall@10 0.59 · Latency p50 2.8 s · Cost $0.015 / useful result · Errors 0%

### Serper — `/serper`
Google SERP scrape · Search · tier T1
Google SERP across web, images, news, places, maps and scholar — plus webpage scraping.
Entry price: $1.00 → $0.30 / 1k · Latency: ~0.6–1.7s · MCP: community
Choose if: You want the cheapest, fastest Google grounding and will synthesize results yourself.
Avoid if: You need multi-engine coverage, rich-SERP features, or built-in content extraction.
Call: https://api.auxiliar.ai/serper/<native-path> — same paths, params and responses as Serper's own API.
Docs: https://docs.auxiliar.ai/serper · Eval page: https://auxiliar.ai/tools/serper/
Measured evals (curator fleet, shared public corpora):
- SERP verticals — composite 9.56/10, rank #1 of 4 (measured 2026-06-22): Quality 1.00 · Latency p50 1.0 s · Cost $0.0003 / call · Errors 0% — Cheapest + fastest, but Google-only.
- Search — composite 7.24/10, rank #3 of 11 (measured 2026-06-20): Recall@10 0.57 · Latency p50 1.5 s · Cost $0.00053 / useful result · Errors 0%

### Tavily — `/tavily`
Agent-native search index · Search · tier T1
An agent-native search index built for RAG — snippet + relevance score + optional answer.
Entry price: $8 (basic) / $16 (adv) / 1k · Latency: ~1.0s · MCP: official
Choose if: You want the agent/RAG default that returns a relevance score and optional answer in one call.
Avoid if: You need per-result dates or a documented compliance/SLA posture.
Call: https://api.auxiliar.ai/tavily/<native-path> — same paths, params and responses as Tavily's own API.
Docs: https://docs.auxiliar.ai/tavily · Eval page: https://auxiliar.ai/tools/tavily/
Measured evals (curator fleet, shared public corpora):
- Crawl — composite 8.38/10, rank #3 of 5 (measured 2026-06-22): Coverage 0.78 · Latency p50 174 ms · Cost $0.014 / useful · Errors 0% — Strong via its /map variant.
- Answer — composite 7.13/10, rank #4 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.24 · Latency p50 1.4 s · Cost $0.008 / call · Errors 0%
- Search — composite 6.66/10, rank #7 of 11 (measured 2026-06-20): Recall@10 0.58 · Latency p50 2.1 s · Cost $0.014 / useful result · Errors 0%

### You.com — `/youcom`
Own crawl index + Research API · Search · tier T2
An own-index web search API plus a cited Research API, tuned for LLM grounding.
Entry price: $5 / 1k · MCP: official
Choose if: You want an independent index with soft domain-boosting and a cited Research API.
Avoid if: You need a flat result shape (results live under results.web[] with snippets[] — needs an adapter).
Call: https://api.auxiliar.ai/youcom/<native-path> — same paths, params and responses as You.com's own API.
Docs: https://docs.auxiliar.ai/youcom · Eval page: https://auxiliar.ai/tools/youcom/
Measured evals (curator fleet, shared public corpora):
- Answer — composite 7.79/10, rank #2 of 7 (measured 2026-06-21): Correctness 1.00 · Citation faithfulness 0.51 · Latency p50 4.3 s · Cost $0.02 / call · Errors 0%
- Search — composite 7.41/10, rank #2 of 11 (measured 2026-06-20): Recall@10 0.65 · Latency p50 1.4 s · Cost $0.0077 / useful result · Errors 0%

## Scraping & Crawling

### Bright Data — `/brightdata`
Web Unlocker + Scraper API · Scraping & Crawling · tier T1
The strongest anti-bot unlocker (98% on the hardest sites) plus pre-built structured scrapers, with top compliance.
Entry price: ~$1.30–1.50 / 1k · MCP: official
Choose if: You face the hardest anti-bot targets and want the strongest unblocker + compliance posture.
Avoid if: You need crawl link-discovery — its url_collection dataset returns only the seed.
Call: https://api.auxiliar.ai/brightdata/<native-path> — same paths, params and responses as Bright Data's own API.
Docs: https://docs.auxiliar.ai/brightdata · Eval page: https://auxiliar.ai/tools/brightdata/
Measured evals (curator fleet, shared public corpora):
- Scrape — composite 7.82/10, rank #4 of 10 (measured 2026-06-20): Anti-bot bypass 75% · Markdown clean 5.1 / 10 · Latency p50 3.9 s · Cost $0.0019 / success · Errors 0%
- Scrape-domain — composite 6.66/10, rank #6 of 6 (measured 2026-06-22): Accuracy 0.72 · Field fill 0.75 · Latency p50 8.1 s · Cost $0.0035 / useful · Errors 25%
- Crawl — composite 4.47/10, rank #4 of 5 (measured 2026-06-22): Coverage 0.10 · Latency p50 6.5 s · Cost $0.01 / useful · Errors 0% — url_collection dataset returns only the seed — no link discovery.

### Crawlbase — `/crawlbase`
Single-call + managed crawl · Scraping & Crawling · tier T2
Markdown + structured scrapers plus a managed async crawl queue, billed per success by domain complexity.
Entry price: $3 / 1k std (success ×complexity) · MCP: community
Choose if: You want a managed crawl queue and pay-for-success by domain complexity.
Avoid if: You want predictable behavior — two distinct tokens (Normal vs JS) and a polymorphic body field.
Call: https://api.auxiliar.ai/crawlbase/<native-path> — same paths, params and responses as Crawlbase's own API.
Docs: https://docs.auxiliar.ai/crawlbase · Eval page: https://auxiliar.ai/tools/crawlbase/
Measured evals (curator fleet, shared public corpora):
- Scrape-domain — composite 7.17/10, rank #5 of 6 (measured 2026-06-22): Accuracy 0.71 · Field fill 0.88 · Latency p50 4.7 s · Cost $0.0028 / useful · Errors 0% — Non-deterministic on Instagram's anti-bot.
- Screenshot — composite 7.15/10, rank #6 of 8 (measured 2026-06-21): Valid image 0.62 · Full-page 1.00 · Latency p50 3.7 s · Cost $0.0015 / call · Errors 25%
- Scrape — composite 7.08/10, rank #8 of 10 (measured 2026-06-20): Anti-bot bypass 75% · Markdown clean 4.0 / 10 · Latency p50 3.5 s · Cost $0.006 / success · Errors 25%

### Firecrawl — `/firecrawl`
Scrape + crawl + extract · Scraping & Crawling · tier T1
Best-in-class LLM-ready markdown, with schema-driven extraction, async crawl and an official MCP.
Entry price: ~$0.83 / 1k (credit/page) · MCP: official
Choose if: You want the highest-quality scrape/markdown and a single surface for scrape, crawl, extract and screenshot.
Avoid if: You're cost-sensitive on hard targets — stealth + JSON stack ~5–9× the base credit.
Call: https://api.auxiliar.ai/firecrawl/<native-path> — same paths, params and responses as Firecrawl's own API.
Docs: https://docs.auxiliar.ai/firecrawl · Eval page: https://auxiliar.ai/tools/firecrawl/
Measured evals (curator fleet, shared public corpora):
- Scrape — composite 9.81/10, rank #1 of 10 (measured 2026-06-20): Anti-bot bypass 100% · Markdown clean 9.6 / 10 · Latency p50 962 ms · Cost $0.0021 / success · Errors 0% — Best-in-class — clears every anti-bot vendor.
- Screenshot — composite 9.5/10, rank #1 of 8 (measured 2026-06-21): Valid image 1.00 · Full-page 1.00 · Latency p50 1.3 s · Cost $0.001 / call · Errors 0% — Top valid-image rate.
- Parse · PDF/doc — composite 8.82/10, rank #1 of 3 (measured 2026-06-22): Text accuracy 0.91 · Latency p50 293 ms · Cost $0.0011 / useful · Errors 0% — OCRs scanned PDFs — the differentiator.
- Extract · AI/schema — composite 8.81/10, rank #2 of 4 (measured 2026-06-21): Field accuracy 0.96 · Latency p50 1.8 s · Cost $0.0052 / useful · Errors 0% — Hallucinated a null field on the precision test.
- Crawl — composite 8.61/10, rank #1 of 5 (measured 2026-06-22): Coverage 0.92 · Latency p50 16.3 s · Cost $0.024 / useful · Errors 0% — Top on clean path-scoped enumeration.
- Watch — composite 8.33/10, rank #1 of 1 (measured 2026-06-21): Class. accuracy 1.00 · Diff quality 0.50 · Latency p50 3.0 s · Cost $0.001 / call · Errors 0% — The only synchronous change-detection provider.
- Act · declarative — composite 7.41/10, rank #3 of 6 (measured 2026-06-22): Task success 0.80 · Latency p50 4.8 s · Cost $0.005 / call · Errors 0%
- Act · NL-agent — composite 5.66/10, rank #2 of 2 (measured 2026-06-22): Task success 0.67 · Latency p50 70.2 s · Cost $0.05 / call · Errors 0%

### Oxylabs — `/oxylabs`
Scraping API + proxy network · Scraping & Crawling · tier T2
Enterprise web scraping with dedicated parsers, a 195+ country proxy network and pay-for-success.
Entry price: ~$0.80–1.00 / 1k · MCP: official
Choose if: You want enterprise-grade scraping, dedicated Amazon/Google parsers, or the leading NL-agent mode.
Avoid if: You need path-scoped crawl enumeration — its AI-prompt crawler can't do it.
Call: https://api.auxiliar.ai/oxylabs/<native-path> — same paths, params and responses as Oxylabs's own API.
Docs: https://docs.auxiliar.ai/oxylabs · Eval page: https://auxiliar.ai/tools/oxylabs/
Measured evals (curator fleet, shared public corpora):
- Scrape-domain — composite 8.74/10, rank #1 of 6 (measured 2026-06-22): Accuracy 0.93 · Field fill 1.00 · Latency p50 2.8 s · Cost $0.0022 / useful · Errors 0% — Top scraper; official first-party APIs are preferred where they exist.
- Extract · CSS rules — composite 8.52/10, rank #2 of 3 (measured 2026-06-22): Field accuracy 1.00 · Latency p50 9.6 s · Cost $0.0012 / useful · Errors 0%
- Act · NL-agent — composite 7.99/10, rank #1 of 2 (measured 2026-06-22): Task success 1.00 · Latency p50 47.3 s · Cost $0.05 / call · Errors 0% — Leads the autonomous NL-agent mode.
- Scrape — composite 7.1/10, rank #7 of 10 (measured 2026-06-20): Anti-bot bypass 75% · Markdown clean 4.0 / 10 · Latency p50 10.5 s · Cost $0.0013 / success · Errors 0%
- Crawl — composite 3.18/10, rank #5 of 5 (measured 2026-06-22): Coverage 0.27 · Latency p50 35.5 s · Cost $0.13 / useful · Errors 50% — AI-prompt crawler can't do path-scoped enumeration.

### ScraperAPI — `/scraperapi`
Single-call scraping API · Scraping & Crawling · tier T1
General-purpose scraping with rotating proxies and auto-retry, behind one GET.
Entry price: ~$0.49 / 1k (credit ×mult 1→75) · MCP: none
Choose if: You want the simplest possible unblocking call and don't need MCP-native discovery.
Avoid if: You need cost transparency (credit multipliers bite) or a first-party MCP.
Call: https://api.auxiliar.ai/scraperapi/<native-path> — same paths, params and responses as ScraperAPI's own API.
Docs: https://docs.auxiliar.ai/scraperapi · Eval page: https://auxiliar.ai/tools/scraperapi/
Measured evals (curator fleet, shared public corpora):
- Screenshot — composite 7.78/10, rank #5 of 8 (measured 2026-06-21): Valid image 0.88 · Full-page 1.00 · Latency p50 27.4 s · Cost $0.0049 / call · Errors 12%
- Scrape-domain — composite 7.26/10, rank #4 of 6 (measured 2026-06-22): Accuracy 0.73 · Field fill 0.78 · Latency p50 2.3 s · Cost $0.0067 / useful · Errors 0%
- Scrape — composite 6.53/10, rank #10 of 10 (measured 2026-06-20): Anti-bot bypass 62% · Markdown clean 4.4 / 10 · Latency p50 2.4 s · Cost $0.01 / success · Errors 38%

### Scrapfly — `/scrapfly`
Single-call scraping API · Scraping & Crawling · tier T1
Strong one-flag anti-bot (asp=true), markdown/AI output and an in-response cost breakdown.
Entry price: ~$3.37–4.13 / 1k · MCP: official
Choose if: You want benchmark-leading anti-bot bypass plus the highest AI-extraction accuracy.
Avoid if: You need crawling or credit rollover — hard targets run ~30 cr/req.
Call: https://api.auxiliar.ai/scrapfly/<native-path> — same paths, params and responses as Scrapfly's own API.
Docs: https://docs.auxiliar.ai/scrapfly · Eval page: https://auxiliar.ai/tools/scrapfly/
Measured evals (curator fleet, shared public corpora):
- Scrape — composite 9.48/10, rank #2 of 10 (measured 2026-06-20): Anti-bot bypass 100% · Markdown clean 9.6 / 10 · Latency p50 4.5 s · Cost $0.0025 / success · Errors 0% — Clears every anti-bot vendor (one-flag asp=true).
- Extract · AI/schema — composite 8.96/10, rank #1 of 4 (measured 2026-06-21): Field accuracy 1.00 · Latency p50 4.5 s · Cost $0.002 / useful · Errors 0% — Highest extraction accuracy in the set.
- Screenshot — composite 8.56/10, rank #4 of 8 (measured 2026-06-21): Valid image 1.00 · Full-page 1.00 · Latency p50 12.0 s · Cost $0.002 / call · Errors 0%
- Act · declarative — composite 7.01/10, rank #4 of 6 (measured 2026-06-22): Task success 0.80 · Latency p50 17.4 s · Cost $0.002 / call · Errors 0%

### ScrapingBee — `/scrapingbee`
Single-call scraping API · Scraping & Crawling · tier T2
One REST surface bundling JS rendering, residential/stealth proxies, AI + CSS extraction and markdown.
Entry price: ~$0.98 / 1k (JS) · MCP: official
Choose if: You want one endpoint for rendering, stealth, extraction and declarative actions.
Avoid if: You're cost-sensitive — JS-on-by-default (5 cr) and stealth (75 cr) escalate fast.
Call: https://api.auxiliar.ai/scrapingbee/<native-path> — same paths, params and responses as ScrapingBee's own API.
Docs: https://docs.auxiliar.ai/scrapingbee · Eval page: https://auxiliar.ai/tools/scrapingbee/
Measured evals (curator fleet, shared public corpora):
- Extract · CSS rules — composite 9.23/10, rank #1 of 3 (measured 2026-06-22): Field accuracy 1.00 · Latency p50 2.7 s · Cost $0.00098 / useful · Errors 0% — Best rule-based extractor — far cheaper than AI extract.
- Act · declarative — composite 8.72/10, rank #1 of 6 (measured 2026-06-22): Task success 1.00 · Latency p50 5.5 s · Cost $0.006 / call · Errors 0% — Best on declarative browser actions.
- Screenshot — composite 8.57/10, rank #3 of 8 (measured 2026-06-21): Valid image 0.88 · Full-page 1.00 · Latency p50 2.7 s · Cost $0.0028 / call · Errors 12%
- Scrape-domain — composite 8.19/10, rank #3 of 6 (measured 2026-06-22): Accuracy 0.90 · Field fill 0.95 · Latency p50 3.6 s · Cost $0.011 / useful · Errors 0%
- Extract · AI/schema — composite 8.14/10, rank #3 of 4 (measured 2026-06-21): Field accuracy 0.89 · Latency p50 3.8 s · Cost $0.0067 / useful · Errors 0%
- Scrape — composite 7.45/10, rank #5 of 10 (measured 2026-06-20): Anti-bot bypass 88% · Markdown clean 4.1 / 10 · Latency p50 8.9 s · Cost $0.0057 / success · Errors 0%

### Spider — `/spider`
Crawl / AI-focused platform · Scraping & Crawling · tier T2
Fast recursive crawl and clean markdown with transparent per-page USD costs in the response.
Entry price: ~$0.03 / 1k (usage-based) · MCP: official (22 tools)
Choose if: You want fast crawling, clean markdown and transparent per-page pricing (anti-multiplier).
Avoid if: You need declarative actions (clicks no-op live) or forecastable usage-based billing.
Call: https://api.auxiliar.ai/spider/<native-path> — same paths, params and responses as Spider's own API.
Docs: https://docs.auxiliar.ai/spider · Eval page: https://auxiliar.ai/tools/spider/
Measured evals (curator fleet, shared public corpora):
- Crawl — composite 8.54/10, rank #2 of 5 (measured 2026-06-22): Coverage 0.77 · Latency p50 2.9 s · Cost $6.2e-05 / useful · Errors 0% — Fast recursive crawl.
- Parse · PDF/doc — composite 6.99/10, rank #3 of 3 (measured 2026-06-22): Text accuracy 0.60 · Latency p50 2.0 s · Cost $3.8e-05 / useful · Errors 0% — Returns raw bytes on scanned PDFs.
- Scrape — composite 6.73/10, rank #9 of 10 (measured 2026-06-20): Anti-bot bypass 50% · Markdown clean 9.6 / 10 · Latency p50 12.1 s · Cost $0.00013 / success · Errors 0%
- Screenshot — composite 5.03/10, rank #7 of 8 (measured 2026-06-21): Valid image 0.50 · Full-page 0.50 · Latency p50 8.4 s · Cost $0.001 / call · Errors 38%
- Act · declarative — composite 1.5/10, rank #6 of 6 (measured 2026-06-22): Task success 0.00 · Latency p50 35.1 s · Cost $0.001 / call · Errors 0% — Declarative clicks no-op live.

### ZenRows — `/zenrows`
Scraper API + CDP browser · Scraping & Crawling · tier T2
A scraper API plus a CDP scraping browser on one key, with heavy anti-bot behind a single flag.
Entry price: ~$1.40 / 1k (JS) · MCP: official
Choose if: You want one key spanning a scraper API and a full scraping browser, with pay-for-success.
Avoid if: You hit sites it blocklists — it errored heavily on *.toscrape.com in the corpus.
Call: https://api.auxiliar.ai/zenrows/<native-path> — same paths, params and responses as ZenRows's own API.
Docs: https://docs.auxiliar.ai/zenrows · Eval page: https://auxiliar.ai/tools/zenrows/
Measured evals (curator fleet, shared public corpora):
- Scrape — composite 7.31/10, rank #6 of 10 (measured 2026-06-20): Anti-bot bypass 62% · Markdown clean 9.4 / 10 · Latency p50 4.8 s · Cost $0.0047 / success · Errors 25%
- Act · declarative — composite 6.89/10, rank #5 of 6 (measured 2026-06-22): Task success 0.80 · Latency p50 10.1 s · Cost $0.005 / call · Errors 0%
- Extract · CSS rules — composite 5.51/10, rank #3 of 3 (measured 2026-06-22): Field accuracy 0.50 · Latency p50 4.5 s · Cost $0.0028 / useful · Errors 50% — Blocklists all *.toscrape.com — an honest gap.
- Screenshot — composite 2.47/10, rank #8 of 8 (measured 2026-06-21): Valid image 0.25 · Latency p50 23.3 s · Cost $0.0014 / call · Errors 75% — Blocked by books.toscrape — an honest gap.

### Zyte — `/zyte`
Single-endpoint extraction · Scraping & Crawling · tier T1
One endpoint with automatic ban-management and built-in AI extraction (product/article/job).
Entry price: $0.13–$1.27 / 1k (HTTP) · MCP: community
Choose if: You want automatic ban-management plus typed AI extraction for product/article/job pages.
Avoid if: You need custom extraction fields — the fixed schema can't honor them.
Call: https://api.auxiliar.ai/zyte/<native-path> — same paths, params and responses as Zyte's own API.
Docs: https://docs.auxiliar.ai/zyte · Eval page: https://auxiliar.ai/tools/zyte/
Measured evals (curator fleet, shared public corpora):
- Screenshot — composite 9.04/10, rank #2 of 8 (measured 2026-06-21): Valid image 1.00 · Full-page 1.00 · Latency p50 5.9 s · Cost $0.001 / call · Errors 0%
- Scrape — composite 8.63/10, rank #3 of 10 (measured 2026-06-20): Anti-bot bypass 100% · Markdown clean 4.2 / 10 · Latency p50 6.5 s · Cost $0.00069 / success · Errors 0% — Clears every anti-bot vendor.
- Act · declarative — composite 8.5/10, rank #2 of 6 (measured 2026-06-22): Task success 1.00 · Latency p50 11.8 s · Cost $0.001 / call · Errors 0%
- Extract · AI/schema — composite 5.7/10, rank #4 of 4 (measured 2026-06-21): Field accuracy 0.55 · Latency p50 6.5 s · Cost $0.0018 / useful · Errors 0% — Fixed schema can't honor custom fields.

## Browser Automation

### Apify — `/apify`
Actor marketplace · Browser Automation · tier T2
A marketplace of 7,000+ Actors for scraping and automation, under one key and bill.
Entry price: varies (CU + rental + result) · MCP: official
Choose if: You want a pre-built Actor for a specific site/task rather than building a scraper.
Avoid if: You need a universal response schema — each Actor emits its own item shape.
Call: https://api.auxiliar.ai/apify/<native-path> — same paths, params and responses as Apify's own API.
Docs: https://docs.auxiliar.ai/apify · Eval page: https://auxiliar.ai/tools/apify/
Measured evals (curator fleet, shared public corpora):
- Scrape-domain — composite 8.23/10, rank #2 of 6 (measured 2026-06-22): Accuracy 0.95 · Field fill 1.00 · Latency p50 8.5 s · Cost $0.0032 / useful · Errors 0%

### Browserbase — `/browserbase`
Browser-as-a-service · Browser Automation · tier T2
Headless browser sessions you drive with Playwright/Puppeteer — for multi-step and auth-walled flows.
Entry price: $0.12 / browser-hr (+ GB) · MCP: official (Stagehand)
Choose if: You need multi-step, stateful or auth-walled flows and want to drive the browser yourself.
Avoid if: You want a one-shot URL→content call — it returns a session handle, not content.
Call: https://api.auxiliar.ai/browserbase/<native-path> — same paths, params and responses as Browserbase's own API.
Docs: https://docs.auxiliar.ai/browserbase · Eval page: https://auxiliar.ai/tools/browserbase/
Measured evals (curator fleet, shared public corpora):
- Act / Screenshot — gated, not scored. Evaluated but gated: Browserbase is a CDP browser session, not a synchronous goal→content REST surface, so it can't be scored on the request/response corpora — it needs a stateful adapter.

## Voice & Audio

### ElevenLabs — `/elevenlabs`
Text-to-speech & voice · Voice & Audio
The full ElevenLabs REST API for text-to-speech and voice, exposed via wildcard routing.
Entry price: see ElevenLabs pricing · MCP: official
Choose if: You need high-quality text-to-speech or voice cloning through the same gateway key.
Call: https://api.auxiliar.ai/elevenlabs/<native-path> — same paths, params and responses as ElevenLabs's own API.
Docs: https://docs.auxiliar.ai/elevenlabs · Eval page: https://auxiliar.ai/tools/elevenlabs/
Not part of the web-access evals (voice provider — the search/scrape/crawl corpora don't apply).

## Pricing
- $5 free trial credit at sign-up, no card required.
- Usage is billed at each provider's real metered price; there is no per-call markup.
- One 5.5% service fee on credit purchases and auto-top-ups ($10–$5,000 per
  purchase). No subscriptions, no monthly fees; an unused provider costs $0.
- Zero balance → requests return HTTP 402 until you top up.
- Details and fee math: https://auxiliar.ai/pricing/

## Reliability & measurement
- Live status page: https://status.auxiliar.ai
- Every provider is benchmarked on the same public task corpora — quality,
  latency p50, cost per success, error rate — and the measured scorecards are
  published with their run dates on each /tools page. Latest run: 2026-06-22.
- Weak scores are published, not hidden. The service fee is flat at top-up, so
  the rankings carry no incentive to steer you toward pricier providers.

## Links
- Docs — base URL, auth, per-provider routes: https://docs.auxiliar.ai
- Docs for machines: https://docs.auxiliar.ai/llms.txt (append .md to any docs page for clean Markdown)
- Tool catalog with measured evals: https://auxiliar.ai/tools/
- Full machine-readable catalog: https://auxiliar.ai/llms-full.txt
- Eval data as JSON (raw metrics, run dates, routes): https://auxiliar.ai/evals.json
- Live status: https://status.auxiliar.ai
- Pricing: https://auxiliar.ai/pricing/
- Sign-up ($5 free): https://accounts.auxiliar.ai/sign-up