Give AI agents structured academic search with PaperSearch MCP
PaperSearch MCP handles academic search, duplicate merging, relevance screening, and topic grouping. Results retain titles, authors, abstracts, years, sources, and DOIs.
Four problems with using general web search for literature research
Intermediate results consume context
As an agent searches and evaluates papers one by one, query traces and irrelevant results keep consuming its context window.
Web results lack citation data
Blogs and secondary summaries may omit authors, years, titles, or DOIs, so they cannot go straight into a bibliography.
The same paper appears repeatedly
A preprint, journal version, and aggregator page may be treated as separate results, creating duplicate review work.
Screening criteria drift
When each search round applies different criteria, the same paper can receive a different inclusion decision.
What happens in one PaperSearch MCP tool call
Search academic sources in parallel
Run queries across multiple academic sources while preserving the provenance of every result.
Deduplicate across sources
Compare DOI, arXiv ID, and normalized titles to merge repeated records.
Screen with consistent criteria
Define screening criteria first, then assess relevance paper by paper. Every result retains its inclusion or exclusion rationale.
Group papers by topic
Organize screened papers by research direction and return structured metadata.
The same search engine already powers Acadwrite Deep Literature Research
The web version provides the same search, duplicate merging, relevance screening, and topic grouping workflow. The MCP version makes these steps callable from MCP-compatible clients.
- Review queries and screening criteria before execution
- Retain the evidence used to merge duplicate records
- Receive titles, authors, abstracts, years, and DOIs
- Keep an inclusion or exclusion rationale for every paper
Frequently asked questions
What is PaperSearch MCP?
PaperSearch MCP is a Model Context Protocol tool for academic literature search. Once connected, an AI agent can call search, duplicate merging, relevance screening, and topic grouping functions.
Why do general-purpose models and coding agents often miss papers when searching directly?
Paper-by-paper searching fills the context with query traces, duplicate records, and irrelevant results. Web pages may also omit authors, years, and DOIs. A reliable search should define its scope and screening criteria before it begins.
How is PaperSearch MCP different from letting an agent search the web?
PaperSearch MCP returns academic literature records and consistently handles duplicate merging, relevance screening, and grouping. General web search may mix papers with blogs, news, and secondary commentary.
Which academic sources does PaperSearch MCP search?
It searches authoritative academic sources including Google Scholar, Semantic Scholar, OpenAlex, Crossref, and arXiv preprints, with support for Chinese and English literature. Multiple queries run concurrently across sources, and results are deduplicated using DOI, platform IDs, normalized titles, and other signals.
Will search results consume a large part of my agent's context?
The agent receives screened and grouped results. Raw query traces and every excluded record are not copied into the conversation context.
Can I use PaperSearch MCP now?
The MCP version is currently accepting waitlist registrations. When access opens, we will send setup documentation and client configuration examples. The web version is available now.
Related tools and guides
Join the PaperSearch MCP waitlist
Leave your email and we will send setup documentation and configuration examples when access opens, with support for MCP clients such as Claude Code and Cursor.
We will use this address only for PaperSearch MCP availability updates, not for other marketing email.