Scry

Scry is a programmatic internet research service built for AI agents and developers who need to compute across many public records rather than receive a conventional page of ranked links. It exposes a large, continuously growing collection of source-native data through an MCP server and an HTTP API. Covered sources include social discussions, academic literature, reference data, public web pages, prediction markets, financial and government records, and software ecosystems. The product's core idea is that an agent can express the research operation it needs, run it against the relevant relations, and receive rows whose identifiers and provenance remain tied to the original source.

An MCP client connects to Scry at a single hosted endpoint and completes an account sign-in. The service works with agent environments that can use remote MCP servers, while other software can call its HTTPS endpoints with an API key. Clients inspect the live schema before constructing a query because each source keeps its own fields, timestamps, coverage, and known gaps. Queries are read-only and operate under explicit limits for time, memory, returned rows, and budget. That structure is useful for research tasks where reproducibility and bounded cost matter as much as getting an answer.

Scry goes beyond keyword lookup. Its query surface supports exact phrases, exclusions, regular expressions, fuzzy and proximity matching, time windows, joins, aggregation, and recursive graph walks. Source-native keys make it possible to follow reply trees, citation networks, authors, market activity, and other relationships without flattening every corpus into one generic document shape. Vector helpers can create named embeddings from text, combine concepts, construct contrast directions, project components, and rank supported corpora by the result. It can also rerank documents an application already holds and expose a web-search endpoint through configured providers.

The service pays unusual attention to provenance and coverage. Result rows retain the fields that identify where a record came from, and schema metadata describes observed extents, lag, and declared holes for each relation. That helps a researcher distinguish a missing row from a source interval the archive may not cover. Data freshness varies by relation and is stated per relation rather than promised globally. Some live sources land within minutes, while snapshots and extracted research corpora advance on their own schedules. Scry also publishes source counts, measured query examples, and the boundaries attached to individual surfaces.

The product fits technical researchers, investigative teams, data journalists, evaluation builders, and agent developers whose questions require whole sets or multi-step computation. A standing query can watch for a precise condition, a graph program can walk a discussion or citation chain, and an aggregate can summarize records without downloading an entire corpus first. It is less suitable for someone who only wants a friendly consumer search box. Users need to understand the schema, choose relevant relations, and write or review the program an agent proposes. Commercial access also sits on a much higher-priced team plan than individual non-commercial research.

Cost and execution controls are part of the query model rather than an afterthought. A caller can limit wall time, computational burden, and financial exposure before a statement runs. Response envelopes report accounting information, and an explain mode can forecast what a query would touch without executing the normal statement. Congestion can change the price of heavy work, while patient and priority behavior support different tolerance for waiting. These controls make sense for automated agents, which might otherwise issue a broad query without appreciating its scope. They also add concepts that users of ordinary web search won't have encountered, so careful clients should surface budgets and partial-result behavior clearly.

Scry uses a freemium structure with metered options. The Researcher plan costs nothing, includes an API key and MCP access, provides five dollars of signup credit without a card, and is limited to non-commercial work. The Patron plan costs one hundred dollars per month, with the payment becoming a rolling balance for non-commercial use. Commercial Team access starts at two thousand dollars per month and adds dedicated capacity and custom source builds. An account-free agent lane costs five cents per declared second through x402, with congestion able to affect pricing. The service is in open alpha, so buyers should expect the available corpora, operators, and interfaces to continue changing.