
Corral
by lamalab-org
Scientific agent benchmarking framework for evaluating AI performance on materials science tasks.
What it does
Corral is a comprehensive benchmarking framework specifically designed for AI agents performing scientific research. It provides standardized environments and metrics to evaluate how well agents handle complex materials science and mathematical challenges.
Tools
- Task Environments: Specialized servers for math, spectroscopy, and molecular dynamics.
- Evaluation Metrics: Automated scoring for pass@k and success rates.
- Checkpoint System: State management for resuming interrupted benchmark runs.
- MCP Integration: Ability to convert scientific tools into MCP format for client use.
Installation
{
"mcpServers": {
"corral": {
"command": "python3",
"args": ["-m", "corral.server"]
}
}
}
Supported hosts
- Claude Desktop
Quick install
pip install corralInformation
- Pricing
- free
- Published
- 5/18/2026
- stars






