OpenAIRE MCP — Connecting AI agents to 150M+ scientific records

Overview

Scientific knowledge is the highest-value corpus an AI agent can draw on — and one of the hardest to reach well. The OpenAIRE Research Graph aggregates 150M+ publications, datasets and software records from European and global open-science infrastructure, with the metadata that makes research navigable: authorship, funding, citations, organisations, projects. Search engines surface fragments of it. AI agents, until recently, saw none of it in a form they could reason over.

At Alien Intelligence we build and operate an MCP (Model Context Protocol) server that exposes the Research Graph to AI agents as a typed tool surface: search and retrieval across research products, author and organisation profiles, project and funding lookups, citation networks, bibliometric impact classes. An agent can go from “find the key papers on this topic” to “map who funds this field in Europe” in a handful of tool calls — every answer traceable to graph records rather than scraped snippets.

My role

I lead the project end to end at Alien Intelligence: the partnership with OpenAIRE, the tool-surface design — what an agent should be able to ask of a research graph, and what it shouldn’t have to know about the underlying APIs — and the path from prototype to operated service.

Where it stands

A first version of the server is operational and in use with AI agents, and we launched an open hackathon with OpenAIRE to put it in the hands of researchers and developers. A full write-up — architecture, tool-surface design choices, and what we learned about agent-shaped access to scholarly data — will follow as the project matures. The same principles drive the Gallica / BnF and LDS / Copyfair work.