Search
One query across six engines — the keyless one works with nothing configured — then pull the top hits through the scrape pipeline in the same call.
On this page
Run a search
- Open Search. Type the query.
- Pick an engine and a result count — 10, 25 or 50.
- Press the search button. Hits come back as rows: position, title, URL, snippet.
The search itself is a single call and writes no run row. Each hit is copyable; a hit fetched through Scrape top opens its page’s Markdown in place.
Engines
| Engine | Key needed | Where the key comes from |
|---|---|---|
duckduckgo |
none | — (the default) |
searxng |
your instance | The SearXNG URL, set in Keys |
brave |
yes | Settings → Keys, the environment, or per search |
serper |
yes | same |
serpapi |
yes | same |
exa |
yes | same |
With nothing configured, the search runs on DuckDuckGo. A keyed engine without a key fails with the engine’s own error plus the exact environment variable to set — MAGPIE_BRAVE_API_KEY and friends — and buttons to save or paste one. A generic MAGPIE_API_KEY never reaches a search engine; each engine takes only its own.
Scrape top
Scrape top — off, 3, 5 or 10 — fetches that many top hits through the normal scrape pipeline: fetch, clean, Markdown, no LLM, selectors included. The page arrives with the hit, in the same call. A hit that can’t be fetched carries the reason verbatim instead — a challenge, a timeout — and the rest still come back.
Every fetched hit is a normal scrape run: it lands in History with its own run id, its page counts against the week’s tiles, and its content checks into the snapshot history so the URL is diffable in Compare later.
Keys and costs
The Advanced field takes an API key for the picked engine, for this search only — it overrides a saved key and is never stored. A search with Scrape top on costs no LLM tokens; the fetches are plain scrapes. One successful search counts once against the trial, fetched hits or not.
The same capability exists for agents: the MCP server’s search tool takes the same engines and an optional scrape-through, and the CLI spells it magpie search "query" --scrape-top 3 (CLI reference).