Materials Project

The Materials Project is a structured scientific database, not a text corpus. It holds computed properties for well over 100,000 inorganic materials, worked out using density functional theory, a physics method for modelling how electrons behave in a solid. Each entry can include formation energy and thermodynamic stability, electronic band structure and band gaps, elastic and magnetic properties, simulated X-ray diffraction patterns, and phase diagrams, and the catalogue deliberately includes many predicted compounds that have never been made in a lab.

Access runs through a free API once you register for a key. The official Python client, mp-api from the pymatgen ecosystem, returns clean, typed records, so you can request exactly the materials and fields you want and serialise them to JSON rather than scraping web pages. That makes it easy to feed straight into your own processing.

Because the data is numeric and relational, the real RAG work is turning records into retrievable text. Most builders template each material into a readable summary, composition, key properties, and stability, embed that, and keep the raw numbers alongside for exact lookup. It suits materials-science assistants, property-lookup tools, and scientific copilots that reason about candidate compounds.

Keep one caveat front of mind: these are computed, not measured, values. Density functional theory has known systematic errors, most famously underestimating band gaps, so present the numbers as high-quality predictions rather than experimental fact, and have your system say so. Coverage is strongest for inorganic crystalline materials, not polymers or messy organics.

The licence is CC BY 4.0, so commercial use is fine as long as you credit the project. Among comparable databases such as OQMD, AFLOW, and NOMAD, the Materials Project is usually the most approachable entry point thanks to its polished API and documentation.