Access any page.
Even the hardest ones.
String is a web data API for developers and AI agents that fetches, searches and maps any public site without getting blocked. One HTTP call against any URL returns clean content, and you only pay when it comes back.
Fast
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
It gets through the systems that others can't, and recovers on its own when a target shifts.
Every request escalates automatically: lightweight request-based fetches, full browser rendering, premium proxies, geolocated routing, and CAPTCHA solving - which compliance-sensitive teams can switch off per request.
AI application companies
Give your agents live web data. In the September 16, 2026 benchmark, String got through more often and faster than Firecrawl and Browserbase.
Want your users to query live web data from inside your product? Connect the API or the MCP server.
GTM teams
Enrichment: Build agents that pull live web data for any deal or contact.
A Claygent-style research agent that doesn't get blocked.
Data companies
Quality: Reach pages you couldn't before because of anti-bot walls or engineering-capacity limits.
Stale data and coverage gaps turn into customer tickets.
Vibe coders
Ship without a scraping stack: point a script at any page and get clean data back, no proxies, retries, or bot walls to babysit.
Building a side project or an agent on a weekend? Pull the pages you need with a single call instead of losing Saturday to rotating proxies and a Cloudflare challenge that wasn't there yesterday.
The Web Data Frontier Benchmark, in full
Every provider ran against every target at the same time. Two numbers decide whether a scraper is worth running: did it get through, and how fast.
We tested String Web Access API against fifteen other web access APIs by loading 100 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, social platforms, and similarly protected surfaces. Each URL was requested 5 times per provider - 500 requests each - and a request only counted as successful when the API returned a 2xx and the response contained the page's expected content.
Latency includes every target. Content uses the 75th percentile of successful attempt latencies. A complete miss uses the 75th percentile of providers that returned content for that target, or the 90-second timeout when nobody succeeds.
One call. That's the integration.
Send a URL, get clean content back. No proxy list to rotate, no headless browser to babysit.
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, Perplexity, or anything that speaks the protocol.
Already on Apify? The same engine ships as 30 ready-to-run Actors on the Apify Store. See the list.
Frequently asked questions
What is the Web Access API?
A web scraping API for humans and agents. One HTTP call against any URL returns the page as clean content, fully rendered and past whatever anti-bot is sitting in front of it. You only pay when it comes back.
Which anti-bot systems does it get through?
Cloudflare, DataDome, PerimeterX, Akamai, Kasada, and CAPTCHA challenges. Every request escalates on its own: a lightweight fetch first, then full browser rendering, premium proxies, and geolocated routing if the target needs it. CAPTCHA solving is part of the same escalation, and teams with compliance constraints can turn it off per request.
How fast is it?
Browser pools stay warm and the rendering path skips a cold spin-up wherever a target allows it, so a burst of requests doesn't turn into a queue.
How reliable is it?
As of October 2026, String passed 98.6% of requests in the open-source Web Data Frontier Benchmark (https://usestring.ai/benchmark), the best success rate of any provider in the October 2026 run. The run covers 100 hard targets at 500 requests per provider, and repeats monthly.
How many requests can I run at once?
Up to 10,000 fetches and 5,000 browser sessions at once, on every plan. The request rate is what grows with your plan, and higher rates are available on request.
Can my agent use it?
Yes. Add String over MCP and your agent fetches any page as a single tool call, in Claude, ChatGPT, Cursor, Perplexity, or anything that speaks the protocol.
How much does it cost?
Start with 20,000 free standard fetches, no card needed. After that, plans start at $19/month, and the fee comes back as usage credit billed per request. Blocked requests cost nothing. Prices as of October 2026, from the String pricing page (https://usestring.ai/pricing).