Free Hybrid Academic Search for AI Agents

Cito is a free academic search engine and developer service that searches more than 236 million papers through keyword and semantic retrieval.

It also provides a JSON API and a remote MCP endpoint to connect the same academic index to Claude Code, Codex, Cursor, and other AI agents.

Visit Cito

Cito serves the retrieval stage of a research workflow. It returns ranked papers with titles, authors, abstracts, publication dates, citation counts, DOIs, and open-access PDF links. Your AI agent or app handles reading, evaluation, synthesis, and writing.

Semantic Scholar’s public API caps anonymous requests at one per second. An AI agent running a deep-research pass fires dozens of queries in sequence and stalls against that limit. Cito serves the same underlying corpus from its own local indexes on a single CPU box. No query in the search path calls the upstream API, and no agent run stalls against a shared rate limit.

How Cito Searches Academic Papers

Every hybrid query runs through two retrieval systems.

The semantic retriever embeds the query with SPECTER2 and compares it against precomputed paper vectors. This route finds conceptually related work even when the title or abstract uses different terminology.

The keyword retriever uses BM25 across paper titles and abstracts. It preserves exact phrases, technical vocabulary, author terminology, and uncommon acronyms.

Reciprocal-rank fusion merges the two rankings. A cross-encoder then reranks the strongest candidates. Queries run against Cito’s own hosted indexes, so upstream academic API rate limits do not interrupt the search path.

Data Coverage, Freshness, and Licensing

Cito currently indexes 236,464,100 papers for keyword search, and 146 million of those carry SPECTER2 embeddings for semantic search.

The corpus comes from Semantic Scholar, which provides metadata, abstracts, citation counts, and the embeddings themselves, and from OpenAlex, which provides open-access PDF links and abstract backfill linked by DOI.

Coverage varies by field: 93.8 million papers carry an abstract (40%), 54.9 million carry a direct open-access PDF link (23%), and 141 million carry a DOI (60%). Medicine, engineering, and environmental science account for the largest shares of the corpus by subject label.

The corpus syncs to each Semantic Scholar bulk release within a day, through a nightly job that enriches new papers with OpenAlex open-access data in the same pass.

Semantic Scholar’s own bulk releases trail its live API by up to a week. A new paper typically becomes searchable in Cito one to two weeks after it lands in Semantic Scholar. Same-day preprint coverage requires a direct query to the preprint server.

Cito holds no full text. The pdf_url field points to the copy at its original host, publisher or repository. Cito never proxies or re-hosts that file. The copyright and access terms of a linked PDF belong to its original host.

Semantic Scholar’s corpus licenses under ODC-BY 1.0. That license permits commercial use and redistribution and requires keeping the attribution field, “Data from Semantic Scholar,” that every API response carries. OpenAlex data is CC0. The SPECTER2 embeddings and model license under Apache-2.0. Commercial products, research, and publications can use Cito results under these terms.

Key Features

Searches more than 236 million academic records.
Combines BM25 keyword retrieval with SPECTER2 semantic matching.
Opens browser search with no account.
Returns structured JSON through a standard HTTP API.
Connects to MCP clients through a remote streamable HTTP endpoint.
Includes abstracts, citation counts, DOIs, venues, and publication dates when available.
Links directly to legal open-access PDF copies when the corpus contains one.
Supports author and publication-date filters in the web interface.
Exposes a leak-safe publication cutoff through the API.
Identifies the corpus release used for each search response.

How to Use Cito

Search from the Website

Open the Cito website.
Select Hybrid, Semantic, or Keyword mode.
Enter a research topic, method name, acronym, or author query.
Apply author or publication-date filters when the topic needs a narrower scope.
Review the paper title, abstract, authors, venue, citation count, DOI, and PDF link.
Open the publisher, repository, or Semantic Scholar record for further reading.

Search Through the JSON API

The API base URL is:

https://api.cito.fim.ai

A basic search request needs only the q parameter:

curl "https://api.cito.fim.ai/search?q=contrastive+learning&limit=5"

Add an API key through either supported header:

curl "https://api.cito.fim.ai/search?q=contrastive+learning&limit=5" \
  --header "Authorization: Bearer sk_..."

The X-API-Key header works as an alternative:

curl "https://api.cito.fim.ai/search?q=contrastive+learning&limit=5" \
  --header "X-API-Key: sk_..."

A search hit contains fields such as:

{
  "id": "13756489",
  "title": "Attention is All you Need",
  "authors": ["Ashish Vaswani"],
  "year": 2017,
  "publication_date": "2017-06-12",
  "venue": "NeurIPS",
  "doi": "...",
  "url": "...",
  "pdf_url": "...",
  "cites": 100000,
  "score": 0.97
}

Search parameters:

Parameter	Purpose
`q`	Sets the natural-language or keyword query
`mode`	Selects `hybrid`, `semantic`, or `keyword`
`limit`	Returns up to 100 results
`enrich`	Adds locally stored paper metadata
`published_before`	Excludes papers that cannot prove publication before the cutoff
`include_imprecise`	Keeps undated or imprecisely dated records with a cutoff

Connect Cito to Claude Code via MCP

Add the remote MCP server with the following command:

claude mcp add --transport http cito https://api.cito.fim.ai/mcp

Add an API key for the authenticated rate tier:

claude mcp add --transport http cito https://api.cito.fim.ai/mcp \
  --header "Authorization: Bearer sk_..."

The endpoint exposes two MCP tools:

search_papers searches the indexed corpus.
get_paper retrieves one paper’s metadata by CorpusID.

Cito API Endpoints

Endpoint	Purpose
`GET /search`	Searches papers and returns ranked hits
`GET /paper/{corpusid}`	Returns metadata for one CorpusID
`GET /status`	Reports index counts, corpus release, and service features
`GET /me`	Returns the authenticated account
`GET /me/usage`	Returns recent authenticated usage

Alternatives & Related Resources

Pros

No signup for web or API search
15x Semantic Scholar’s anonymous rate limit
Native MCP endpoint installs with one command
Deterministic, replayable result ordering
Commercial use permitted under the data licenses

Cons

New preprints take 1-2 weeks to appear
No full text; PDF links depend on OpenAlex
Abstract coverage sits at 40% of the corpus
Retrieval only; no summarization or synthesis
Anonymous limit: 15 requests per minute

FAQs

Q: Does Cito search Google Scholar?
A: No. Cito searches its own locally mirrored indexes built from Semantic Scholar data and OpenAlex enrichment.

Q: Can Cito results be used in a commercial application?
A: The underlying metadata and embeddings permit commercial use under their respective licenses. Applications that display or redistribute results must preserve the Data from Semantic Scholar attribution. Rights for linked PDF files follow the terms of the publisher or repository that hosts each document.

Q: Can Cito return full-text PDFs?
A: No. Cito returns a direct link to the PDF at its original host when OpenAlex has recorded an open-access copy. It never stores or proxies the file itself.

Q: Can I filter by publication year?
A: Yes. On the website use Advanced search to set a date range. In the API, the published_before parameter excludes papers published on or after a given date, and you can combine it with published_after by crafting your own client-side filter.

Q: Is there a paid tier for higher rate limits?
A: Currently no. The default API key limit is 100 requests/min. If you need more, contact the creator.