Yahia Bakour has a theory about what's holding back AI agents: they're starving for fresh web data, and most developers are either building fragile scrapers themselves or paying enterprise rates for infrastructure that wasn't designed with language models in mind.
His New York-based startup, Context.dev, emerged recently with what might charitably be called an unglamorous pitch. Instead of positioning itself as the next revolution in AI reasoning or multimodal search, the company is selling plumbing—a unified API that replaces the messy stack of web scraping tools, brand data vendors, and screenshot utilities that AI builders currently cobble together. Backed by Y Combinator, Context.dev is betting that as agentic workflows proliferate, someone needs to solve the boring infrastructure problem first.
The homepage frames it plainly: "the web context API for teams building software and AI agents." One endpoint scrapes URLs into clean Markdown or HTML. Another crawls entire sites. A third pulls typed brand data—logos, colors, fonts, social handles, industry classifications. A fourth enriches those cryptic transaction strings from bank statements (the "SQ *..." or "AMZN MKTP..." variety). All of it designed to feed structured data directly into agent pipelines without the usual headaches: rate limits, anti-bot detection, stale cached responses.
It's infrastructure, which is to say it's not sexy. But infrastructure has a way of mattering.
What's Under the Hood
Context.dev offers four product categories, according to its documentation. Web scraping endpoints convert any URL into GitHub-flavored Markdown, raw HTML, or structured image lists. A site crawl API returns each page as Markdown—useful if you're building a knowledge base or training a chatbot on a company's entire web presence. Screenshot endpoints capture specific pages (pricing, terms of service, contact forms) with page-type heuristics baked in. And brand intelligence APIs retrieve company data by domain, name, email, stock ticker, or even messy transaction descriptor.
The performance specs tell part of the story. According to the company's documentation, cold requests target a median latency of around seven seconds, with the 90th percentile at 18 seconds and the 99th percentile under a minute. Cached hits return in under a second. Anti-bot stealth is included at no surcharge—a pointed contrast to competitors that layer on fees for JavaScript rendering or residential proxies. Failed or blocked requests aren't billed, which is a small mercy but not an insignificant one when you're debugging agent workflows at scale.
Integration spans TypeScript, Python, Ruby, Go, and PHP SDKs, plus REST. There's also a Model Context Protocol server for Claude and Cursor, and no-code connectors for Make, Zapier, and Excel. The MCP server, in particular, signals where Bakour sees the market headed: agents that query the live web directly, not just pre-indexed datasets or static knowledge bases.
Whether that vision plays out is anyone's guess. But the bet is clear.
The Reliability Pitch
The company's status page shows uptime snapshots spanning several months: HTML scraping above 99.9%, image extraction near five nines, sitemap parsing similarly robust. Context.dev moved to credit-based pricing in March 2026, with web scraping endpoints costing one credit per call or page, screenshots at five credits, and brand retrieval at ten credits. Rate-limit improvements rolled out over the summer, adding proper Retry-After headers and ensuring throttled requests never hit users' bills—a detail that matters more than it sounds.
The free tier offers 500 credits and 30 requests per minute for work emails, dropping to 250 credits and 10 requests per minute for personal accounts. Paid plans start at $49 monthly for 30,000 credits, scaling to enterprise volumes with 99.9% SLAs and custom rate limits. There's also a separate logo CDN (branded "Logo Link") that sits outside API quotas and isn't rate-limited, useful for pulling brand assets at high frequency without burning through your credit balance.
Early Customers, Real Use Cases
The company's Y Combinator profile states they are "proud to power agentic products at 250 other companies." Customers include Klarna, Mintlify, Super, Vizzy, Similarweb, and a handful of developer-focused startups.
Case studies offer specifics, which is rare enough in startup land to be worth noting. SiteGPT, which builds AI chatbots trained on entire websites, migrated from competitor Firecrawl in under a day according to their published case study, citing cost and scale constraints. Mintlify, a docs automation tool, integrated Context.dev's brand APIs in ten minutes to automatically style documentation sites with company colors, fonts, and logos pulled from GitHub repos. The speed of integration—ten minutes—strains credulity slightly, but Mintlify confirmed the timeline in a case study published on Context.dev's site.
Sim.ai, an agent orchestration platform, lists Context.dev as an official integration, letting users chain web scraping and brand lookups inside multi-step workflows. A Show HN post drew attention to the anti-bot features and Markdown-first design, aligning with broader conversations around productionizing agents—a theme that's moved from conference panels to actual infrastructure decisions, perhaps more quickly than anyone expected.
Who's Behind It
Bakour spent time as a software engineer at Amazon and Sunrun before co-founding StockAlarm.io, a stock tracking app that scaled to over 225,000 users before being acquired, according to his LinkedIn and the Y Combinator directory. He graduated from Boston University in 2020, which makes him relatively young to be tackling infrastructure plays, though not unusually so in the YC ecosystem.
Context.dev's LinkedIn company page lists the team size at two to ten employees—a wide band that suggests either deliberate opacity or the kind of flux common to very early startups. Recent posts reference infrastructure scaling and a surge of overnight signups, though without hard numbers it's difficult to gauge momentum. No separate funding amount or valuation has been disclosed beyond the standard YC investment. The legal entity is Bakour Tech LLC, doing business as Context.dev, registered in Delaware.
Context.dev rebranded from Brand.dev earlier this year to reflect an expanded scope—moving beyond brand data into web scraping, structured extraction, and AI-powered queries. The changelog notes the shift happened before a major rollout of scraping endpoints, with backward compatibility maintained. A deprecated search API suggests the company is narrowing focus to scraping and brand enrichment rather than generic web search, which is probably wise given how crowded that space has become.
Open Questions
The platform is live and generally available. Enterprise customers get SSO, SAML, SCIM, and dedicated support alongside custom volume commitments. The trust center shows SOC 2 Type I compliance complete and Type II in progress, with subprocessors including Cloudflare, Google Cloud, and Hetzner. Terms of service clarify that Context.dev facilitates access to web content but doesn't grant licenses to use it—a standard API platform position, though one that leaves downstream liability with the customer.
For developers tired of stitching together web scrapers, proxy services, and brand data vendors, Context.dev offers a consolidated alternative with agent-first design. Whether that's enough to carve out durable market share against entrenched competitors like Bright Data, Apify, and Firecrawl remains an open question. Those companies have years of infrastructure maturity, established enterprise relationships, and battle-tested anti-detection systems. Context.dev is betting on superior developer experience and pricing transparency, which are real advantages but not necessarily defensible ones.
Still, timing matters. Agent workflows are multiplying across industries, and the refrain that "agents are only as good as the context we give them" has become something of a rallying cry among AI infrastructure builders. If Context.dev can execute on reliability and keep onboarding friction low, there's a plausible path to becoming the default plumbing for a new generation of agentic products.
Plausible. Not guaranteed. But in infrastructure, plausible is often enough to start.
