Web Access API

Access any page.
Even the hardest ones.

One HTTP call against any URL. String returns clean content, and you only pay when it comes back.

Get an API key ↗Read the docs

First request in seconds · Free to start · Pay only on success

Live · fetch

$ curl https://request.usestring.ai/v1/fetch \

-H "Authorization: Bearer sk_live_…" \

-d '{ "url": "https://www.zillow.com/homedetails/…" }'

200 · 412ms · 18.4 KB clean text

<!doctype html> … fully rendered, anti-bot cleared, no proxy to manage.

Fast

0.41s

p50 latency

0.9s

p95 latency

Warm browser pools and a rendering path that skips a cold spin-up wherever a target allows it. Concurrency holds the tail flat under load, so a burst of requests doesn't turn into a queue.

Reliable

Reliable means it gets through the systems that others can't, and recovers on its own when a target shifts.

CloudflareDataDomePerimeterXAkamaiKasadaCAPTCHA challenges

Every request escalates automatically: lightweight request-based fetches, full browser rendering, residential proxies, geolocated routing, and CAPTCHA solving - which compliance-sensitive teams can switch off per request.

Q2 2026 · re-run monthly

The Web Data Frontier Benchmark, in full.

Every provider, every target, same time, head to head. The two numbers that decide whether a scraper is worth running: did it get through, and how fast.

53 hard targets14 providers265 requests per providerdata generated June 29, 2026View the open-source benchmark on GitHub ↗

Benchmark methodology

Real pages, real defenses, verified content.

We tested String Web Access API against thirteen other web access APIs by loading 53 real-world pages known to be difficult to scrape, including targets protected by anti-bot and WAF systems such as DataDome, Akamai, Cloudflare, and Kasada.

The suite covers pages with high scraping value: product detail pages, search results, and similarly protected commerce surfaces. Each URL was requested 5 times per provider - 265 requests each - and a request only counted as successful when the API returned a 2xx and the response contained the page's expected content.

01 · Overall success rate

All targets, all requests. Higher is better.

string

97.4%

bright

79.2%

zenrows

76.2%

context_dev

75.5%

scraperapi

73.2%

firecrawl

72.1%

decodo

69.4%

oxylabs

69.1%

zyte

67.5%

nimble

58.5%

browserbase

50.6%

scrapingdog

47.9%

scrapingbee

35.1%

scrapingant

32.8%

0%25%50%75%100%

02 · Average latency

Mean per-request latency across all attempts. Lower is better - read it alongside success rate.

string

11.00s

bright

24.14s

zenrows

23.47s

context_dev

13.69s

scraperapi

17.85s

firecrawl

12.82s

decodo

32.97s

oxylabs

20.89s

zyte

18.22s

nimble

31.18s

browserbase

2.03s

scrapingdog

10.17s

scrapingbee

5.83s

scrapingant

6.86s

0s9s18s27s36s

Get an API key ↗Free to start · Pay only on success

Data generated June 29, 2026 (Q2 2026 run).

One call. That's the integration.

Send a URL, get clean content back. No proxy list to rotate, no headless browser to babysit.

Request · curl

curl https://request.usestring.ai/v1/fetch \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{ "url": "https://hard-target.com", "format": "markdown" }'

200 · 412ms · 18.4 KB clean text

Give your agents the same reach.

Add String over MCP and your agent fetches any page, even the hardest ones, as a single tool call, in Claude, ChatGPT, Cursor, or anything that speaks the protocol.

MCP · Claude / ChatGPT / Cursor

{
  "mcpServers": {
    "string": {
      "url": "https://mcp.usestring.ai/v1/mcp",
      "headers": { "Authorization": "Bearer sk_live_…" }
    }
  }
}

Get your API key now.

Get an API key ↗