news_pipeline, news_branches, news_events, and news_articles.
What the pipeline produces
Events
Clusters of related articles grouped around a single real-world occurrence — an earnings release, a regulatory action, an M&A deal. Each event has a label, a structured summary, and an article count.
Articles
Individual news items with a relevance score (1–10) and a bucket classification. Articles scored above the relevance threshold are bucketed as
relevant; the rest go to discard.Branches
Every pipeline run operates against a branch — a named configuration that controls what gets fetched, how articles are scored, and when the pipeline runs automatically.Production branch
The production branch runs on a schedule automatically. Its results are what you see by default in
news_events and news_articles. You cannot delete it.Custom branches
Create your own branch to experiment with different fetch sources (
fetch_config) or a custom scoring prompt (extraction_prompt). Results stay isolated from production until you are satisfied.When you call
news_events or news_articles without specifying a branch_id, results come from the production branch.MCP tools at a glance
Event types
The pipeline classifies each article into one of the following event types during the extraction step:Next steps
Running the pipeline
Run the full pipeline or individual steps, and check run status.
Managing branches
Create and configure custom branches with your own fetch sources and scoring prompts.