Most market data was designed for humans to read, then retrofitted for machines to parse. Agents deserve the opposite: intelligence that is structured, judged, and cited at the source. That's what Forecite ships.
Verdicts are structured judgement
An agent consuming raw headlines inherits the hardest problem in the pipeline — deciding what matters. Forecite events arrive with that decision made and itemized: actionability with five sub-scores, direction and conviction, tags across four dimensions, and the symbols involved. Your agent branches on fields, not on a paragraph it has to interpret.
Plug in where the agent lives
If your agent runs in Claude, Cursor, or anything else that speaks MCP, the Forecite MCP server gives it tools to search the feed, pull an article with its full verdict breakdown, and score arbitrary text — one config block, entitlement-aware, same limits as your API key.
Building your own harness instead? The same surfaces are plain HTTP: the WebSocket feed with server-side filtering (an agent subscribed to 'US biotech, actionable only' receives exactly that), the REST API for history, and signed webhooks when the agent shouldn't hold a socket.
Score what the agent finds
Agents don't just consume your feed — they discover documents in the wild: a PDF, a regulatory page, a rumor on a forum. The scoring endpoint accepts any text and returns the same verdict schema as the feed, so 'found it myself' and 'came off the wire' flow through one decision path.
Everything is metered per key with explicit rate-limit headers, so a fleet of agents degrades predictably instead of mysteriously. Start with one tool call; the first verdict tends to make the argument.