# @kalei-labs/mcp-server

MCP server for [KALEI](https://kaleiai.com) — AI cognitive profiling tools for Claude Code, Cursor, and any [Model Context Protocol](https://modelcontextprotocol.io) client.

## What it does

Exposes 10 KALEI tools to your AI assistant:

- `kalei_leaderboard` — Top models by Cognum score
- `kalei_model` — Detailed profile of any model
- `kalei_compare` — Head-to-head comparison
- `kalei_dimensions` — The 10 cognitive dimensions
- `kalei_providers` — Supported BYOK providers
- `kalei_profile_model` — Profile any model (BYOK)
- `kalei_run_progress` — Track ongoing runs
- `kalei_run_result` — Get final profiles
- `kalei_insights` — Auto-generated findings
- `kalei_ask_parliament` — 4-model deliberation

## Quick start

### 1. Get a KALEI API key
[kaleiai.com/get-started](https://kaleiai.com/get-started) → sign up → Dashboard → API Keys

### 2. Configure your MCP client

**Claude Code** (`~/.claude.json` or settings):
```json
{
  "mcpServers": {
    "kalei": {
      "command": "npx",
      "args": ["-y", "@kalei-labs/mcp-server"],
      "env": {
        "KALEI_API_KEY": "kalei_live_..."
      }
    }
  }
}
```

**Cursor** (`.cursor/mcp.json`):
```json
{
  "mcpServers": {
    "kalei": {
      "command": "npx",
      "args": ["-y", "@kalei-labs/mcp-server"],
      "env": { "KALEI_API_KEY": "kalei_live_..." }
    }
  }
}
```

### 3. Use it

In Claude Code / Cursor chat:

> Show me the KALEI leaderboard and compare Claude Sonnet vs GPT-5.

> Profile gpt-4o for me using my OpenAI key sk-... — use standard depth.

> Ask the Parliament: should AI systems have persistent memory?

The model will call the appropriate tools and return results.

## Development

```bash
git clone https://github.com/LMGameLabs/kalei-mcp-server
cd kalei-mcp-server
npm install
npm run build
```

## Environment variables

- `KALEI_API_KEY` (required) — your KALEI API key
- `KALEI_BASE_URL` (optional) — override default `https://kaleiai.com/api/v1`

## License

MIT — LM Game Labs