Summary
The three web primitives an agent keeps reaching for, without three separate integrations.
read_webpage and search_web are the daily pair, and the switches that decide whether output is usable — images, links, generated alt text, cache bypass — are exposed rather than assumed on your behalf. fact_check is the outlier: it takes a statement rather than a query, which makes it right for checking a claim you already have in hand and wrong for open-ended research.
What it is
A three-tool front end for Jina AI: page extraction, web search, and statement grounding.
What you get
- A URL turned into content shaped for a model, with links and images kept or dropped and generated alt text on request
- no_cache on the page read, for when you need the page as it stands right now rather than as it was
- Web search returning a chosen count of results in the return_format you ask for
- A statement checked against Jina's grounding engine, with deepdive for a longer pass
Requirements
JINA_API_KEY. All three tools call Jina's API, so the key is the only thing between the server and an answer — there is nothing local to install and nothing to configure per site.
Setup effort
One command plus a key — npx -y jina-ai-mcp-server, then supply credentials