The summaries come back long enough to reason over rather than as a one-line snippet, which is the difference between search a model can answer from and search it has to click through. The cards cover the questions where the answer changes by the minute — a share price or today's weather is rarely reliable on whatever page a query surfaces.
A search server from Bocha with two tools. Web search returns page titles, links, summaries, publication times and site names from a very large index. AI search adds structured cards on top, selected from what the query means — weather, encyclopaedia, calendar, trains, stocks and dozens more.
- Web results with title, link, summary, publication time and site name
- A time window on the search — a specific date, a date range, or the last day, week, month or year
- Control over how many results come back
- Structured cards alongside the results for domains where a page cannot carry the current value: weather, stocks, exchange rates, fuel prices, trains, almanac, medical and encyclopaedia among others
- Coverage that reaches beyond web pages into partner content — video, news, encyclopaedia, travel, hospitality and academic sources
An API key from the Bocha open platform, supplied as an environment setting. The server runs as a Python package launched with uv from a clone of the repository.
One command plus a key — uvx bocha-search-mcp, then supply credentials