hindcast

MCP servers

MCP server for earnings call transcripts

An earnings call is an hour of speech in which perhaps ninety seconds matter, and which ninety seconds depends entirely on the question being asked. Feeding an entire transcript into a model to find them is expensive, lossy and unreliable — the relevant exchange is often a hedged answer to the fourth analyst rather than anything in the prepared remarks.

This server exposes calls as searchable passages. An agent queries a company and a topic, and receives the specific exchanges with the speaker, their role and the timestamp within the call. Prepared remarks and Q&A are separated, because a rehearsed sentence and an unscripted answer to a pointed question are not equivalent evidence.

Every passage returns with enough identity to be quoted responsibly: which call, who spoke, and when in the call it happened.

hindcast — session

preview

hindcast> did management commit to a margin target on the call?

→ tool: transcripts_search { entity:"...", type:"earnings_call", query:"margin" }

matching exchanges with speaker, role and in-call timestamp

source: call transcript, prepared remarks and Q&A separated

Preview of the shape of an answer — endpoints open to the waitlist first.

What's covered

Join the waitlist — this page's dataset is already selected.

Need this one?votes for Global Media Transcripts
Optional — helps us build in the right order. All skippable.

Role

Company

Datasets you need

Questions

Why passages instead of whole transcripts?

Because context windows are finite and relevance is narrow. Returning the exchange that answers the question costs less and produces better answers than returning an hour of text and hoping the model finds the right paragraph.

Can an agent compare what was said across quarters?

That is a natural use of the tool — the same query across several calls returns comparable passages, which is how a change in language becomes visible without reading every transcript.

Join the waitlist. Vote the roadmap. First in, first served.

← Back to the data catalog