Skip to main content

Search

Search finds articles across your corpus by keyword, topic, or phrase. It uses PostgreSQL full-text search, so queries and article contents stay inside your instance.

Indexed fields​

The search index includes title, summary, reason, tags, source name, and, when article body extraction is enabled, extracted body text.

Results are ranked by PostgreSQL relevance and can include articles in any workflow state: Today, Later, Done, Skipped, Starred, Snoozed, or Archived.

Open search from the navigation or focus it with /. Results show the title, summary snippet, source, state, and publication date. Select a result to open the article in the main view.

If you want to narrow results to a specific state, open that view first and use the search control there.

Searching from an MCP client​

The read-only MCP server exposes the same user-scoped article search through search_news. Use list_news_sources first to discover the subscribed, enabled source slugs available to your token. Both tools require the search scope and use the user identified by the MCP bearer token; there is no user or owner argument.

Source discovery is ordered by category, descending priority, name, then slug. Request 1–25 sources per page (default 25), omitting cursor for the first page. Pass each next_cursor back unchanged until it is null. The cursor is a canonical ASCII-decimal string of at most 20 digits; a cursor at or beyond the end returns an empty terminal page. Source pages use {sources, truncated, next_cursor}. Here, truncated: true means the 4,800-byte budget ended the page early, and next_cursor resumes at the first source not returned.

Each discovered source has slug, name, category, and kind. Exact slug and category values are valid search_news filters. Sources with invalid filter-valued slugs or categories are omitted. Display-only name and kind are capped at 120 characters and may be shortened further when JSON escaping requires it to stay within the 4,800-byte budget.

search_news supports a query plus source, category, date-range, workflow-state, starred, and archive filters. An empty query returns the filtered recent listing in the web search's canonical order. Multiple values within a source, category, or state filter combine with OR; different filter groups combine with AND.

Date ranges use article discovery time, not publication time: day, week, and month cover the trailing 1, 7, and 30 days. Archived articles are omitted unless include_archived is true or archived is explicitly selected as a state. Workflow state and starred status belong to the authenticated token owner, so one user's state cannot leak into another user's results.

Page search results with limit (1–25, default 10) and numeric offset (0–10,000, default 0); this is separate from the source-discovery cursor. Queries are limited to 2,000 characters. Source and category filters accept up to 50 non-empty values of at most 120 characters, and states accept up to 50 values from today, later, done, skipped, and archived.

Search returns {articles, truncated}. Each article includes compact metadata, its canonical URL when available, a summary, and your state and star value; article bodies are omitted. Article responses keep only complete records within the structured-content size budget, so truncated: true means the next complete article did not fit. See Configuration → MCP server for setup and the full argument reference.

Full-text extraction​

Deeper body search depends on optional article body extraction. When enabled, the app fetches and caches article text, then indexes it alongside the normal metadata fields. This increases storage and ingestion work, so it is kept optional.