The whole surface is a pair of searches, each taking a single keyword — no date range, no author, no category. That makes it a first look rather than a research instrument: you find out which names keep appearing, then go elsewhere for the papers themselves. One caution the tool description states outright: arXiv returns best-effort matches for any query, so results coming back is not on its own evidence that prior work exists.
A small literature-search server: one keyword search against arXiv, one against Google Scholar, and an optional web search when a key for it is present.
- arXiv searched by keyword for related articles, with no key needed and an answer in about a second
- Google Scholar searched by keyword through the scholarly library and its free proxy pool — the same question, against reviewed literature
- A Google web search registered as a third tool when SERPBASE_API_KEY is set, and absent otherwise
- An empty keyword comes back as a tool error rather than an empty result set, so a failed search is never mistaken for "nothing has been published"
Uv on the machine and nothing else — no account or key for the two search tools. Both take a keyword and nothing more: there is no date range, author or category argument to narrow with. Allow more time for the Scholar side than the arXiv side; it goes through a free proxy pool and is far less predictable. MIT.
One command — uvx mcp-scholarly