Cito

Product Introduction

1. Definition: Cito is a specialized academic search engine and API that performs hybrid semantic and keyword search over the Semantic Scholar corpus. It is a technical tool in the categories of academic information retrieval, semantic search, and AI agent tooling.
2. Core Value Proposition: Cito exists to provide fast, unlimited, and meaning-based search for academic literature, specifically solving the problem of API rate limits that hinder AI agents and researchers from conducting deep, automated literature reviews.

Main Features

1. Hybrid Search Fusion: Cito combines multiple search methodologies into a single ranked result. It performs a traditional keyword search over 236 million papers, a dense vector similarity search using SPECTER2 embeddings over 146 million papers, fuses the results using Reciprocal Rank Fusion (RRF), and finally reranks them with a cross-encoder model for precision.
2. Developer-First API & MCP Endpoint: The product offers a plain JSON API for programmatic access and a native Model Context Protocol (MCP) endpoint. This allows AI agents like Claude Code to directly query the academic corpus as a tool, bypassing typical upstream rate limits from other academic APIs.
3. Free Web Search with No Signup: Cito provides an immediate-use web interface for semantic paper search without requiring user registration, lowering the barrier to entry for students, researchers, and curious individuals.

Problems Solved

1. Pain Point: AI agents and automated research workflows are severely throttled by the strict rate limits of traditional academic APIs (e.g., Semantic Scholar, PubMed), causing them to fail or become impractically slow during deep literature research.
2. Target Audience: Primary users include AI/ML engineers building research agents, data scientists conducting systematic literature reviews, academic researchers, and developers integrating scholarly search into applications.
3. Use Cases: Essential for automating systematic literature reviews, powering AI research assistants (like Claude Code), building academic discovery tools, and enabling applications that require high-volume, semantic querying of scientific papers without infrastructure constraints.

Unique Advantages

1. Differentiation: Unlike standalone semantic search tools or limited public APIs, Cito is a unified, production-ready system that merges scale (236M papers), modern retrieval methods (hybrid search), and unfettered access (no rate limits for agents). It contrasts with traditional keyword-only search on academic portals.
2. Key Innovation: The specific integration of the SPECTER2 embedding model for dense retrieval, combined with RRF fusion and cross-encoder reranking, applied at scale and exposed via an MCP endpoint. This makes state-of-the-art semantic search directly actionable by AI agents.

Frequently Asked Questions (FAQ)

1. What is Cito semantic search? Cito semantic search is an academic paper search engine that finds relevant research based on the conceptual meaning of your query using SPECTER2 AI embeddings, going beyond simple keyword matching.
2. How does Cito search work? Cito works by performing parallel keyword and vector similarity searches over millions of papers, intelligently combining the results with Reciprocal Rank Fusion, and then reranking the final list with a cross-encoder model for the most relevant outcomes.
3. Is Cito search free to use? Yes, Cito offers a completely free web search interface with no signup required, and provides a free JSON API and MCP endpoint for developers and AI agents.
4. What is the Cito MCP endpoint for? The Cito MCP (Model Context Protocol) endpoint allows AI coding agents like Claude Code to directly search and retrieve academic papers as part of their workflow, effectively giving them deep literature research capabilities without hitting rate limits.
5. What is SPECTER2 in Cito? SPECTER2 is the specific transformer-based AI model used by Cito to generate dense vector embeddings of scientific papers, enabling the semantic "search by meaning" functionality over its 146-million-paper index.