Use cases & endpoints
Financial data for Claude via MCP — connect and query
Claude can already reason about financial questions; what it cannot do unaided is look up what a company reported in a filing three years ago and cite the document. MCP closes that gap by letting the model call tools directly, and a remote MCP server means there is nothing to install locally.
Hindcast is that server for market data. Connect it and Claude gains tools for filings and fundamentals, macro series with vintages, holdings, central bank text, transcripts and entity resolution — all sharing one identifier space, so a follow-up question about the same company does not need to re-establish which company it is.
The design choice worth knowing about is that historical tools take an as-of date. That makes it structurally hard for the model to answer a question about the past using a number published later, which is the failure mode that makes AI financial answers untrustworthy in the first place. Endpoints open to the waitlist first.
hindcast — session
previewhindcast> ask about a past filing and get a citable answer
→ tool: xbrl_facts { ticker:"...", concept:"...", as_of:"..." }
✓ the as-reported value with its filing reference
source: preview — endpoints open to the waitlist first
What's covered
- Remote MCP server — a URL, not a local install
- Tools spanning filings, fundamentals, macro, holdings and text
- As-of arguments on historical tools by default
- Citations Claude can quote rather than paraphrase
- One identifier space across every dataset
- Same endpoint works for IDE agents and production pipelines
Join the waitlist for early access.
Questions
What is MCP, briefly?
An open protocol that lets an assistant call external tools it discovers at connect time. Practically, it means a data provider can expose capabilities once and any compatible client can use them without a bespoke integration on either side.
Does connecting it change how the model reasons?
It changes what the model can check. Instead of recalling a plausible figure, it can retrieve the reported one and show where it came from — which matters most on exactly the questions where a confident guess is most damaging.
Related
Join the waitlist. Vote the roadmap. First in, first served.