# HEA-Bench

> HEA-Bench is an open-source MCP server for health and science. Compute high-entropy alloy and oxide descriptors and phase rules from any composition. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** Web, Windows, CLI, API
- **Repository:** https://github.com/dfieser/hea-bench
- **MCP Registry name:** io.github.dfieser/hea-bench
- **Categories:** [health and science](https://ai-powered.com/categories/health-and-science)
- **Listing updated:** 2026-11-18

## About

Built around the equimolar Cantor alloy CoCrFeMnNi as its sanity check, hea-bench reports descriptors such as mixing entropy, atomic-size mismatch, Miedema mixing enthalpy, valence-electron concentration, Yang-Zhang omega and electronegativity mismatch. Each is a closed-form expression over a curated element table, and the report carries the literature source of every input.

On top of the descriptors it applies the classic empirical phase rules (Yeh, Zhang, Guo-Liu, Yang-Zhang, King, Ye, Senkov-Miracle, the Tsai sigma-phase window), and it covers high-entropy oxides with formability checks for rock-salt, perovskite, fluorite and pyrochlore structures. Fitted predictions for hardness and phase sets carry calibrated uncertainty. Alloy coverage is 55 elements, with Miedema pairs for 75 elements; the one missing pair, Th-U, is reported rather than zeroed.

The MCP server exposes 24 tools that mirror the Python library, a browser app and a portable Windows desktop app, which the README says share one calculation core and are kept in parity by tests.

## Features

- **Alloy descriptors:** Mixing entropy, size mismatch, melting temperature, Miedema enthalpy, VEC, omega and more.
- **Phase-prediction rules:** Applies the standard empirical rules to a composition and returns the predicted phase.
- **High-entropy oxides:** Formability descriptors for rock-salt, perovskite, fluorite and pyrochlore with automatic charge balancing.
- **Fitted predictions with uncertainty:** Hardness and phase prediction sets flag alloys unlike the training data.
- **Per-value provenance:** Reports cite the literature source of each input and carry a content-hashed result ID.
- **Four equivalent surfaces:** Web app, desktop exe, Python library and MCP server cover the same features.

## Good for

- Screen a candidate alloy composition before committing to synthesis
- Let an AI assistant check phase rules for a list of compositions
- Evaluate whether an oxide composition is likely to form a given structure
- Reproduce descriptor values with tabulated references for a paper

## Install

### Claude Code (Terminal)

```sh
claude mcp add hea-bench -- uvx hea-bench
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "hea-bench": {
      "command": "uvx",
      "args": [
        "hea-bench"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "hea-bench": {
      "command": "uvx",
      "args": [
        "hea-bench"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "hea-bench": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "hea-bench"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "hea-bench": {
      "command": "uvx",
      "args": [
        "hea-bench"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "hea-bench": {
      "command": "uvx",
      "args": [
        "hea-bench"
      ]
    }
  }
}
```

## FAQ

### How do I install the MCP server?

Run pip install "hea-bench[mcp]" and start hea-bench-mcp.

### How many elements are supported?

55 elements for alloys; the oxide module's Shannon table covers 94 elements.

### Is there a way to try it without installing?

Yes, a browser app runs entirely client-side at the project's GitHub Pages site.

### Are the rule outputs reliable predictions?

The README calls them simple empirical surrogates, so treat them as fast screens rather than predictions.

### Is HEA-Bench free?

Yes. HEA-Bench is open source, so you can use and self-host it for free.

### How do I add HEA-Bench to Claude Code?

Run this command in your terminal: claude mcp add hea-bench -- uvx hea-bench

### What are the best alternatives to HEA-Bench?

The closest alternatives to HEA-Bench are Tooluniverse, Oya Browser, BasalFit, Ebi Ols and Clinicaltables.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Tooluniverse](https://ai-powered.com/mcp/tooluniverse): Tooluniverse is an open-source MCP server for health and science.
- [Oya Browser](https://ai-powered.com/mcp/oya-browser): Oya Browser is a freemium MCP server for browser automation.
- [BasalFit](https://ai-powered.com/mcp/basalfit): BasalFit is a free MCP server for health and science.
- [Ebi Ols](https://ai-powered.com/mcp/ebi-ols): Ebi Ols is a freemium MCP server for health and science.
- [Clinicaltables](https://ai-powered.com/mcp/clinicaltables): Clinicaltables is a freemium MCP server for health and science.
- [Complex Portal](https://ai-powered.com/mcp/complex-portal): Complex Portal is a freemium MCP server for health and science.
- [Idigbio](https://ai-powered.com/mcp/idigbio): Idigbio is a freemium MCP server for health and science.
- [Ensembl](https://ai-powered.com/mcp/ensembl): Ensembl is a freemium MCP server for health and science.

---
Source: https://ai-powered.com/mcp/hea-bench
