Datasets
SEC & EDGAR data for AI agents (API + MCP)
EDGAR behaves like a filing cabinet, not a database. Full-text search covers a rolling window, XBRL facts arrive in company-specific taxonomies, and the same accounting concept can sit under three different tags across two filers. Getting from a question like what did this company report for revenue in the quarter ending June 2019 to a number an agent can quote usually means writing a scraper, a tag mapper and a cache before any analysis starts.
Hindcast is building that layer once. Filings and their exhibits are parsed and chunked for retrieval, XBRL facts are normalized onto a common concept set, and every value carries the accession number and filing timestamp it came from. As-reported values stay as reported: when a company restates, the restatement is stored as a new observation rather than an overwrite, so a query with an as-of date returns what the market could actually see on that date.
The interface is a remote MCP server for agents that call tools directly, plus REST with OpenAPI for production pipelines. Access opens to the waitlist first, and the build order follows the votes on the catalog.
hindcast — session
previewhindcast> revenue AAPL as reported in the Q3 2019 10-Q — point-in-time
→ tool: xbrl_facts { ticker:"AAPL", concept:"Revenues", as_of:"2019-07-31" }
✓ 53,809,000,000 USD (as-reported, never restated)
source: SEC EDGAR 10-Q filed 2019-07-31 · CIK 0000320193
What's covered
- 10-K, 10-Q and 8-K full text with section anchors and exhibit links
- XBRL company facts normalized across taxonomy versions and filer quirks
- 13F-HR holdings with issuer resolution to CUSIP and ticker
- Form 4 insider transactions with reporting-owner relationships
- Accession number, CIK and filing timestamp attached to every value
- Restatements kept as separate observations, never overwriting the original
Join the waitlist — this page's dataset is already selected.
Questions
How is this different from calling EDGAR directly?
EDGAR gives you documents; this gives you resolved values with provenance. The difference shows up in the boring parts: taxonomy drift between filers, amended filings that supersede originals, issuers that change CIK, and the rate limits that make a naive scraper unreliable in production.
What happens when a company restates a prior period?
Both versions are kept. The original stays queryable at its original filing date and the restated value is added with its own filing date, so an as-of query never silently returns a number that was published later than the date you asked about.
Do I need an SDK to use it?
No. An MCP-capable agent can call the tools directly, and anything else can use plain HTTP against the REST endpoints. Endpoints open to the waitlist first.
Related
Join the waitlist. Vote the roadmap. First in, first served.