Undermind: AI Literature Search on Semantic Scholar

Undermind is an AI research tool that searches the scientific literature for one question and writes a report with citations. Its FAQ says that it draws on the Semantic Scholar database of over 200 million articles. Undermind documents no crawler of its own, so a paper reaches Undermind through Semantic Scholar.

How Undermind searches

The Undermind website describes the search as an agent. It reads and evaluates hundreds of papers, follows citation trails, and changes its strategy as it learns about the topic. The FAQ says that a typical search takes about two minutes.

Undermind searches in stages, not all at once. It uses a statistical model to estimate how many relevant papers it has found so far. If the search is not complete, the report says so, and the user can extend the search. The Undermind whitepaper explains the method.

In the report, each statement has an in-line citation that links back to the source paper.

Sources and coverage

The FAQ gives these details:

• The Semantic Scholar database, which contains PubMed, arXiv and many other databases.
• On the free tier, the abstracts and metadata of papers.
• On the Pro tier, also the full texts of nearly all open access and preprint articles.

Undermind also has an MCP server, so other AI assistants and coding agents can use its search. MCP is a standard way to connect tools to AI assistants.

For authors and site owners

• Undermind documents no crawler name, user agent or robots.txt token.
• If Undermind does not find a paper, the FAQ gives two causes. The paper is not in Semantic Scholar, or the search did not finish.
• If you want Undermind to find your papers, make sure that they are in Semantic Scholar. See the Semantic Scholar guide for its sources and for author pages.

For other AI tools that search research papers, see Elicit and Consensus.