Scry

Scry is a search service at scry.io built for agents rather than for people, exposed as an MCP server. The difference from a search API is the unit of work: instead of returning links for an agent to fetch and read one at a time, Scry lets it submit a bounded program that runs across queryable relations holding roughly 164 billion rows drawn from Reddit, Hacker News, arXiv and Wikipedia. A question that would otherwise mean thousands of page fetches becomes one declared query over the corpus. Its pricing is the unusual part, and the company is explicit about why: a single program can ask for more compute than the servers have at that moment, so the rate floats with demand — the busier the machines, the more a heavy query costs. Agents are billed at $0.05 per second declared. The tiers around that are a free Researcher tier with $5 of starting credit for non-commercial use, a Patron tier at $100 a month, also non-commercial, and a Team tier from $2,000 a month for commercial use with dedicated capacity. Commercial use therefore starts at the $2,000 tier, which is the figure to plan against; the cheaper tiers are licensed for research. The service describes itself as an open alpha. It was announced on Hacker News on 18 September 2026.