Point-in-time filings, survivorship-complete base rates, and event signals you can query "as of" any date. Prototype hypotheses in your AI client; ship on the API.
One protocol — bring it to the assistant you already use
The same jobs, without the exports, the tabs, or the unverifiable answer.
Before
Point-in-time reconstruction meant building your own filing-date joins and hoping nothing leaked.
With MCP
Every tool accepts an as-of date and returns what was knowable then — leak-safe by construction, not by discipline.
Before
Base rates computed on surviving tickers — inflated by everything that quietly disappeared.
With MCP
Survivorship-complete populations, with delistings, failures, and discontinued programs in the denominator.
Before
Testing a feature idea took a week of scraping and parsing before the first read.
With MCP
Probe 529 event classifiers and 152M classified filing sentences in plain English, in an afternoon.
Run these today in Claude, ChatGPT, or any MCP client — each returns a cited answer.
Reconstruct BridgeBio's pipeline as it was knowable at the end of 2020.
Which companies filed Item 4.02 non-reliance 8-Ks last quarter?
What fraction of programs targeting TTR have historically succeeded?
Rank my watchlist by consecutive quarters of net institutional inflow.
Explore hypotheses in Claude against the same point-in-time corpora your production code would use — then ship on the Kaleidoscope API platform. Same data, same provenance contract, production delivery.
No — MCP is the research channel: the fastest way to explore the data and pressure-test ideas. For production pipelines, the same corpora are available through the Kaleidoscope API platform. Ask about delivery options during your demo.
Quarterly XBRL fundamentals back to ~2009 for ~4,400 companies; 13F position history 26+ quarters; classified filing sentences 2022–2026; news 2020–2026. Each tool states its own coverage window — no silent extrapolation.
Answers are reconstructed from what was filed and public on the as-of date. If a fact was not knowable then, it is not in the answer — the same contract that makes backtests honest.
13F refreshes every ~4 hours from EDGAR, news daily, and filings flow through as the SEC publishes them.
Read-only tools, no side effects, and your queries are never used to train models. How we handle data
Request a demo and we'll walk through the workflow above with your names, your dates, and your questions.
Request a Demo