For AI & ML Researchers

Production-Ready Materials Datasets — Out of the Box

Stop spending weeks wrangling raw database exports. MatCraft gives you 205,000+ materials with 30+ clean, normalised properties accessible via REST API, CSV, or JSON.

The Problem

Sound familiar?

Raw Database Exports Are a Mess

MP, AFLOW, and JARVIS use different property names, units, and null conventions. Harmonising them for a training set takes weeks.

No Standard Train/Test Splits

Without community-standard splits, benchmark comparisons across papers are meaningless — everyone uses different subsets.

Property Coverage is Uneven

You need band gap, formation energy, elastic moduli, and magnetic moment — but no single database has all four for the same set of materials.

The Solution

How MatCraft fixes it

Normalised, Multi-Source Data

205,000+ entries with consistent property names, SI units, and explicit nulls. Ready to load into PyTorch or TensorFlow.

30+ Properties per Material

Band gap, formation energy, elastic moduli, density, crystal system, space group, and more — all in one response payload.

API + Bulk Export

Pull individual materials via REST API or export filtered datasets as CSV or JSON for offline training.

Key Features

Built for AI & ML Researchers

REST API

Documented endpoints with Python and cURL examples. Returns JSON with full property payloads. Paginated bulk access supported.

30+ Normalised Properties

Band gap, formation energy, bulk modulus, shear modulus, Poisson ratio, density, magnetic moment, and more.

CSV / JSON Export

Filter by any property combination and export the resulting dataset. Ideal for training GNNs and transformers.

Versioned Data

Track which database version your training set came from. Reproducible benchmarks require reproducible data sources.

We trained our graph neural network on a MatCraft export and got our first publishable results in a week. Normally the data pipeline alone takes a month.

D

Dr. Priya Menon

Postdoctoral Researcher, ML for Materials · Stanford SUNCAT Centre

Pricing built for ai & ml researchers

Pick the plan that matches your workflow

Real Stripe-live prices. Every credit pack is valid 12 months; every subscription rolls 30 days and cancels any time. Stripe-hosted checkout — we never see a card number.

Free tier

Free

$0forever
10 signup credits

Benchmark a proof-of-concept GNN on a real dataset before asking for budget.

  • Materials search + 30+ normalised properties
  • Up to 1,000 rows per CSV export
  • Versioned data (DB snapshot hash on every export)
  • Public Python + cURL examples
Start free
Recommended for youMonthly subscription

Researcher

$49/mo
$0.98 / credit
50 credits / month

Modal tier for a postdoc or PhD training a property predictor. $49 beats the GPU-hour bill of rerunning a MP pull.

  • REST API (60 req/min) + paginated bulk access
  • JSON + CSV exports up to 50k rows
  • Train/test split helpers on request
  • Cancel any time — roll-over 30 days
Start Researcher →
sku: researcher_monthly
Monthly subscription

Professional

$149/mo
$0.75 / credit
200 credits / mo + bulk API

When your benchmark requires the full 205k corpus — or your lab is running 3+ predictors against our data.

  • API (600 req/min) + parallel workers
  • Full-database exports (205k rows)
  • Versioned data with reproducible commit hashes
  • Priority support (4h SLA)
Try Professional →
sku: professional_monthly
Stripe-hosted (PCI SAQ-A) Credits grant via idempotent webhook Cancel any time, no lock-inCompare all 7 SKUs →

Ready to get started?

Join thousands of researchers and engineers already using MatCraft to accelerate materials discovery.