The research firehose problem

Paperzilla is a research discovery platform designed to filter the increasing volume of academic preprints into usable intelligence. Based in Den Haag, the company was founded by Mark Pors, who previously served as CTO of the monitoring service WatchMouse. The platform operates on a fundamental premise: researchers do not need more papers; they need better signals. By monitoring over 10,000 daily uploads across repositories including arXiv, bioRxiv, medRxiv, and ChemRxiv, Paperzilla attempts to solve the discovery problem through a combination of semantic matching and automated classification.

Architecture for humans and agents

The technical architecture is built for both human readers and automated systems. For human users, it provides email digests, RSS feeds, and in-app feeds that rank papers by relevance—categorizing them as "must-read," "related," or "not relevant." This categorization is handled by a precision ranking system that uses LLMs to distinguish between groundbreaking work and tangential studies. For automated systems, Paperzilla provides a Model Context Protocol (MCP) server and a command-line interface (pz CLI). This dual-track approach reflects a shift in how research is conducted: increasingly, LLM agents are the primary consumers of academic literature, performing the initial screening before presenting a shortlist to a human principal.

Feed-first discovery

Most research tools focus on the search box—the reactive discovery of information. Paperzilla focuses on the feed—the proactive delivery of information. This distinction is important because it changes the researcher's relationship with the literature from active hunting to managed monitoring. The platform uses a "two-pass" RAG-backed (Retrieval-Augmented Generation) summarization system that grounds claims in specific citations from the text. This process is designed to reduce the hallucination risks inherent in standard LLM summaries, ensuring that a "smart summary" actually reflects the underlying data rather than free-form guesswork.

Market position and pricing

Competitively, Paperzilla occupies a middle ground between broad academic search engines like Google Scholar and AI-native research assistants like Elicit or ResearchGate. While Elicit is built for answering specific questions, Paperzilla is built for tracking specific domains over time. Its pricing model, currently in a waitlist phase, targets individual scholars at the entry-level while offering institutional and enterprise tiers for lab heads and industry R&D departments. These higher tiers include features like "agentic search," which uses parallel LLM instances to perform deep reads and cross-reference claims across multiple papers.

The agentic workflow utility

The platform inclusion of an MCP server is its most forward-looking feature. It positions Paperzilla not as a standalone destination, but as a core utility for the agentic workflow ecosystem. As AI agents gain the ability to use external tools, having a standardized way to pull high-signal research data becomes a significant advantage. By focusing on "high-signal" rather than "high-volume," the company acknowledges the primary bottleneck in modern research: the limited bandwidth of the human mind. The tool aims to protect that bandwidth by acting as a high-fidelity filter between the firehose of global research and the individuals who need to act on it.