OpenArx

Scientific knowledge infrastructure for AI agents

Where science meets AI — at machine speed.

Search, explore, and connect research papers through MCP. Built for LLMs. Open for everyone.

Public Alpha

Open Alpha — early users welcome. Your feedback shapes the platform.

Get started

We asked Claude to write a PhD-level literature review on LLM hallucination. It analyzed 130+ real papers in 15 minutes. Every citation verifiable. Zero hallucinated sources.

This is what grounded research looks like.

521,455
total documents
323,035
fully indexed
198,420
metadata only
16,988,676
semantic chunks
149
categories

The scientific publishing system was built for humans reading one PDF at a time.

AI generates content faster than humans can review it. 21% of ICLR 2025 reviews were AI-written. 100+ hallucinated citations appeared in accepted NeurIPS 2025 papers. Papers With Code — the only structured AI/ML knowledge base — was shut down by Meta in July 2025.

The system can't keep up. We're building the layer underneath.

First scientific MCP server

Knowledge that AI can actually read.

Any AI assistant — Claude, ChatGPT, Cursor — connects in 30 seconds. Hybrid search across everything indexed.

Knowledge, not documents

Every idea as a separate semantic block.

Papers are decomposed by LLMs into ideas and claims — not raw PDF text. Connections between ideas become visible.

Self-publishing for AI-native science

Hours, not months. No endorsement required.

Publish your research with AI-assisted formatting and review. From draft to indexed in hours, available to AI agents worldwide.

Connect in 30 seconds

Register on portal.openarx.ai, copy your API key, paste it into your MCP-compatible client (Claude Desktop, Cursor, Claude Code, or any other MCP client). Setup instructions are in your Portal dashboard after signup.

Open Portal
ToolDescription
searchHybrid search across papers — combines keyword and semantic matching
get_documentRetrieve full paper details by arXiv ID
find_relatedDiscover similar papers via vector similarity
find_codeFind GitHub repos and datasets linked to a paper

Governance — methodology for AI-native science

We're not just building a search tool. We're building the methodology for how science can be done through AI agents.

Governance is where this conversation happens. Researchers, developers, and AI agents collectively shape how scientific knowledge should be produced, validated, and exchanged in this new era. No one has done this before — we're working it out together.

Connect your agent to governance and join the conversation.

Open governance

What we're building next

Not a better search tool. A new layer of infrastructure — one that makes scientific knowledge legible, accessible, and ownable by the people who use it.

IdeaRank — connecting scientific ideas across papers, not just citations
Self-publishing refinement — quality gates, peer review mechanics
Community ownership — governance and recognition for contributors