# AIRAS

Automates machine learning workflows from academic paper review to final publication. Pre-registers hypotheses in Git and validates numerical metrics via CI.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · ★ 37
- Author: airas-org
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/airas-org/airas
- Updated: 2026-10-16
- HTML: https://affpapa.org/en/nevibecode/mcp/airas

## What it does

The server gives your coding assistant specialized tools for running automated ML studies. The agent searches literature, outlines falsifiable hypotheses, drafts experiment scripts using Hydra, and schedules test runs through GitHub Actions or the Seyval compute platform.

Every experiment step logs directly to `.research/record.json`. Reported figures link into LaTeX manuscripts with the `\airasval{...}` macro, preventing manual data entry errors. The workflow mandates preregistration: paper drafts and criteria commit to Git before runs dispatch, while CLI utilities like `airas verify-record` confirm consistency in CI.

Running the package requires `uv` on your path and a `GH_PERSONAL_ACCESS_TOKEN` stored in `~/.airas/credentials.json` with `repo` and `workflow` scopes. Without GitHub credentials configured, the assistant cannot create and configure the study repository.

## Requirements
- Needs: Python and uv (commands run via uvx)

## Install

### Claude Code (native platform)

**Platform CLI**
```bash
claude mcp add airas -- uvx airas
```
uvx runs the command named after the package; if the README names another one, use uvx --from airas <command>

### Codex

**Platform CLI**
```bash
codex mcp add airas -- uvx airas
```
uvx runs the command named after the package; if the README names another one, use uvx --from airas <command>

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "airas": {
            "command": "uvx",
            "args": [
                "airas"
            ]
        }
    }
}
```
uvx runs the command named after the package; if the README names another one, use uvx --from airas <command>

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "airas": {
            "type": "stdio",
            "command": "uvx",
            "args": [
                "airas"
            ]
        }
    }
}
```
uvx runs the command named after the package; if the README names another one, use uvx --from airas <command>

### Gemini CLI

**Platform CLI**
```bash
gemini mcp add airas uvx airas
```
uvx runs the command named after the package; if the README names another one, use uvx --from airas <command>

## When to use it
- automate the machine learning research loop from literature survey to paper draft
- commit study hypotheses to git prior to initiating test runs
- insert validated experiment metrics into latex manuscripts automatically
- validate reproducibility of research figures and numbers in continuous integration

## Example prompts
- Search recent publications on sparse attention and prepare the initial experimental design
- Preregister our hypothesis commit and generate the project repository
- Extract the latest test results, build Vega-Lite plots, and insert the metrics into paper.tex

## Similar
- [Zotero MCP](https://affpapa.org/en/nevibecode/mcp/zotero-mcp.md) — MCP server · ★ 5548
- [arxiv-mcp-server](https://affpapa.org/en/nevibecode/mcp/arxiv-mcp-server.md) — MCP server · ★ 3231
- [PaperBanana](https://affpapa.org/en/nevibecode/mcp/paperbanana.md) — MCP server · ★ 2417
- [MATLAB MCP Server](https://affpapa.org/en/nevibecode/mcp/matlab-mcp-server.md) — MCP server · ★ 1716
- [Hex Research](https://affpapa.org/en/nevibecode/mcp/hex-research.md) — MCP server · ★ 583
- [DocsAgent — Zotero, Obsidian & Apple Notes MCP Server](https://affpapa.org/en/nevibecode/mcp/docsagent-zotero-obsidian-apple-notes-mcp-server.md) — MCP server · ★ 578

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md