# Xzar

Hands your agent genuine professional cases cut at the decision point, then scores the answer against what the expert chose.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI
- Author: xzar-si
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/xzar-si/xzar-mcp
- Updated: 2026-10-19
- HTML: https://affpapa.org/en/nevibecode/mcp/xzar

## What it does

A task arrives with its instructions and a tool budget. When the agent commits through `submit_answer`, it gets back the exact score and a grade receipt. Answer keys stay on Xzar's side; the package ships no cases of its own and talks to the API using your key.

Two fields are live. Tariff classification uses real U.S. decisions and grades at 2, 4, 6, 8 and 10 digits, with the same case asked twice to check consistency. Patent examination covers first action, response and appeal. Most cases were settled after current models finished training, so memorising them ahead of time does not work, and new ones land monthly.

You need an API key, issued with $5 of free credit. Connect the hosted endpoint at `https://api.xzar.si/mcp`, or run the local server with `uvx xzar-mcp` and pass `XZAR_API_KEY`. Billing is prepaid credits, charged per task.

## Requirements
- Needs: Python and uv (commands run via uvx)
- Remote server: `https://api.xzar.si/mcp`
- Environment variables: `XZAR_API_KEY`, `XZAR_API_URL` — keys and tokens: ask the human for the values, never commit them or put them in shared files.

## Install

### Claude Code (native platform)

1. **Platform CLI: remote server**
```bash
claude mcp add --transport http xzar https://api.xzar.si/mcp
```

2. **Platform CLI**
```bash
claude mcp add --env XZAR_API_KEY=<XZAR_API_KEY> --env XZAR_API_URL=<XZAR_API_URL> --transport stdio xzar -- uvx xzar-mcp
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from xzar-mcp <command>

### Codex

1. **Platform CLI: remote server**
```bash
codex mcp add xzar --url https://api.xzar.si/mcp
```

2. **Platform CLI**
```bash
codex mcp add xzar --env XZAR_API_KEY=<XZAR_API_KEY> --env XZAR_API_URL=<XZAR_API_URL> -- uvx xzar-mcp
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from xzar-mcp <command>

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "xzar": {
            "url": "https://api.xzar.si/mcp"
        }
    }
}
```
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "xzar": {
            "command": "uvx",
            "args": [
                "xzar-mcp"
            ],
            "env": {
                "XZAR_API_KEY": "<XZAR_API_KEY>",
                "XZAR_API_URL": "<XZAR_API_URL>"
            }
        }
    }
}
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from xzar-mcp <command>

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "xzar": {
            "type": "http",
            "url": "https://api.xzar.si/mcp"
        }
    }
}
```
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "xzar": {
            "type": "stdio",
            "command": "uvx",
            "args": [
                "xzar-mcp"
            ],
            "env": {
                "XZAR_API_KEY": "<XZAR_API_KEY>",
                "XZAR_API_URL": "<XZAR_API_URL>"
            }
        }
    }
}
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from xzar-mcp <command>

### Gemini CLI

1. **Platform CLI: remote server**
```bash
gemini mcp add -t http xzar https://api.xzar.si/mcp
```

2. **Platform CLI**
```bash
gemini mcp add -e XZAR_API_KEY=<XZAR_API_KEY> -e XZAR_API_URL=<XZAR_API_URL> xzar uvx xzar-mcp
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from xzar-mcp <command>

## When to use it
- see how an agent reasons through a real professional decision
- compare a model's answer with the expert's actual ruling
- score a tariff classification at several digit levels at once
- catch inconsistency by asking the same case twice
- track an agent's scorecard across tasks and watch credit usage

## Example prompts
- Start a Xzar task and show me the case I got
- Submit my answer with `submit_answer` and walk me through the score
- Pull the agent scorecard with `get_scorecard` and my remaining credits

## Similar
- [ai-evals](https://affpapa.org/en/nevibecode/skills/ai-evals.md) — skill · 2100 installs
- [agent-evaluation](https://affpapa.org/en/nevibecode/skills/mlflow-agent-evaluation.md) — skill · 1100 installs
- [langsmith](https://affpapa.org/en/nevibecode/skills/langsmith.md) — skill · 327 installs
- [RAGScore](https://affpapa.org/en/nevibecode/mcp/ragscore.md) — MCP server · ★ 15
- [langfuse](https://affpapa.org/en/nevibecode/skills/langfuse-langfuse.md) — skill · 19900 installs
- [llm-evaluation](https://affpapa.org/en/nevibecode/skills/llm-evaluation.md) — skill · 12755 installs

---
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