When Yahia Bakour posted his new startup to Hacker News this summer, he made a pitch that probably resonated with every developer who'd ever tried to feed web data into a large language model: stop duct-taping together three different services just to scrape a single webpage.
Context.dev, which emerged from Y Combinator earlier this year, sells what amounts to a single API call that replaces the usual mess. Point it at any URL, and within seconds it returns clean Markdown or JSON—JavaScript rendered, anti-bot defenses handled, proxies included. No add-ons, no surprise fees for the hard parts. According to Product Hunt, Context.dev was listed as #1 of the day on July 2, and it already lists Cursor and Mintlify as customers in Y Combinator's directory.
The timing makes sense. AI agents need to read the internet, and they need that information structured in ways that don't burn through token budgets. Context.dev offers real-time scraping, site-wide crawling, and what the company calls "brand intelligence"—pulling logos, colors, fonts, even NAICS codes straight from a company's web presence. A monitoring feature rolled out in July lets developers check a page for changes every 10 minutes if they want. The batch API allows submission of up to 25,000 URLs or a crawl as a single asynchronous job.
Bakour, who previously worked on retail checkout technology at Amazon and later served as a principal engineer at Sunrun, has some history with high-volume systems. He co-founded StockAlarm.io, a service that racked up more than 225,000 users and pushed out over 100,000 notifications daily before it was acquired. His Y Combinator profile mentions the exit but not the buyer.
What's notable is how quickly some developers have plugged Context.dev into their workflows. Instant, an AI coding tool, migrated from a competitor called Firecrawl and had the integration running within an hour, according to a case study on Context.dev's site. Doeke Leeuwis, Instant's founder, stated the integration was up and running within an hour. Another customer, Bystreet, ran north of a million scrapes for hotel ownership research without hitting snags, the company claims.
The product has picked up integrations with Cursor, Vercel's Eve assistant, and OpenClaw in recent weeks. Each uses a hosted Model Context Protocol server that exposes 34 typed tools for search, scrape, crawl, and extract operations. Other customers include a handful of Y Combinator batchmates—Drip, Hoplite—and companies like Similarweb, SiteGPT, Construct, and Dagny. Case studies on the company's site span from mid-June through late August.
Developer plans start at $25 a month for 60 API calls per minute. Pro runs $149 (300 requests per minute), and Scale costs $499 (700 RPM). Overages get billed in blocks of 10,000 credits at tiered rates. One credit buys one scrape, and JavaScript rendering plus residential proxy costs are bundled in—no markup, Bakour confirmed on Hacker News. There's a free tier: 500 one-time credits at 30 requests per minute if you sign up with a work email, or 250 credits at 10 RPM for personal addresses.
The company says its Markdown scraping endpoint has held 99.988% uptime over the past month, with 100% for images and crawling, based on its public status page. It achieved SOC 2 Type 1 compliance around May, though Type 2 certification is still in progress.
Context.dev faces competition from Firecrawl, which offers a larger free tier and open-source components. Bright Data prices by the gigabyte for browser scraping. Apify runs a marketplace model with per-result fees. Exa focuses specifically on AI-oriented search. Each has carved out a niche, though the underlying challenge is the same: websites weren't designed to be parsed by machines, and making that work reliably at scale is harder than it looks.
The startup was called Brand.dev until March, when it rebranded and expanded beyond brand extraction into general-purpose web scraping. Bakour announced Y Combinator backing on May 31 on the company blog, though the round size and any other investors haven't been disclosed. LinkedIn pegs the headcount somewhere between two and 10 employees as of late August.
In his Launch HN post—which drew 119 upvotes and 86 comments—Bakour said the company maintains a caching layer to avoid hammering sites repeatedly and honors opt-out requests from webmasters. It's a small nod to the tension that underlies this entire category: the web is open, but scraping it at industrial scale can feel like trespassing, depending on who you ask.
For now, Context.dev is betting that developers building AI agents will pay for reliability and simplicity. One API call, one flat rate, and the web data shows up formatted and ready to feed into a model. Whether that's enough to fend off competitors with deeper pockets or more established reputations remains to be seen. But Bakour has at least solved the problem he set out to fix: no more duct tape.
