Summary
One tool, one decision: how many rounds of searching a question is worth.
The whole server is a single research call whose depth is a parameter, which suits questions where the first page of results is a starting point rather than an answer. Since rounds is yours to set, time and cost scale with what you ask for rather than with anything the server decides. There is nothing else in the surface — no fetch, no cache, no source management — so it sits alongside other tools rather than replacing a browser.
What it is
A single-tool research server that runs repeated web searches around a question, its keywords and its topic.
What you get
- Research that iterates: deep-research takes a question with its keywords and topic and works through the number of rounds you set, rather than returning one result page.
- Depth decided per call: rounds sets how far it goes, and reference carries source material into the pass.
Requirements
No account and no key. The tool searches the web, so the host needs outbound network access.
Setup effort
One command plus a key — npx -y mcp-deep-research, then supply credentials