Xzar
Crash-test your AI agent on real professional decisions, graded against what actually happened.
Install
Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).
claude mcp add --transport http xzar 'https://api.xzar.si/mcp'
{
"mcpServers": {
"xzar": {
"type": "http",
"url": "https://api.xzar.si/mcp"
}
}
}
xzar-mcpclaude mcp add xzar -e 'XZAR_API_KEY=<your xzar_api_key>' -e 'XZAR_API_URL=<XZAR_API_URL>' -- uvx xzar-mcp==0.1.2
{
"mcpServers": {
"xzar": {
"command": "uvx",
"args": [
"xzar-mcp==0.1.2"
],
"env": {
"XZAR_API_KEY": "<your xzar_api_key>",
"XZAR_API_URL": "<XZAR_API_URL>"
}
}
}
}
Environment variables
XZAR_API_KEYrequired secret Your Xzar API key (https://xzar.si/signup.html)XZAR_API_URLOverride the API base URL (default https://api.xzar.si); for local development only
Documentation
From the project's README on GitHub, refreshed weekly. Links open on GitHub.
Xzar MCP server
Crash-test your AI agent on real professional decisions. Xzar gives your agent real cases, cut at the moment of decision, and grades its answer against what the professional actually decided. Most cases were decided after current models were trained, so they can't have been memorized; fresh ones are added monthly.
Live today:
- Tariff classification: real U.S. classification decisions, graded exactly at 2, 4, 6, 8 and 10 digits, with a consistency check (the same case asked twice). Testing a tariff classifier? See https://xzar.si/scorecard
- Patent examination: first action, response and appeal.
More fields are on the way; ask for yours at https://xzar.si.
Setup
- Get an API key (with $5 of free credit) at https://xzar.si/signup.html
- Add the server to your MCP client:
{
"mcpServers": {
"xzar": {
"command": "uvx",
"args": ["xzar-mcp"],
"env": { "XZAR_API_KEY": "your-key" }
}
}
}
Hosted connector (no install)
In Claude or any MCP client that supports remote servers: add https://api.xzar.si/mcp and sign in with your Xzar API key.
Claude plugin
The plugin/ folder is the Xzar plugin for Claude. It bundles the hosted connector with two skills: crash-test, which runs a task end to end and explains the score, and xzar-request, for custom tests, pilots, quotes and training access. In Claude Code:
/plugin install xzar --marketplace xzar-si/xzar-mcp
Tools
| Tool | What it does |
|---|---|
start_task | Start a task: the case, the instructions and the tool budget |
search_library | Search the case's reference library |
read_publication | Read one library item |
search_rules | Search the professional rulebook for the task |
read_rule | Read one rulebook section |
submit_answer | Submit the decision; returns the exact score and the grade receipt |
request_service | Ask for a custom test for your field, a pilot, a quote, or training access; get a reference and next steps |
get_scorecard | Your agent's scorecard across tasks |
get_usage | Credit balance and usage |
Answer keys never leave Xzar's servers. This package holds no tasks; it calls the Xzar API with your key.
Pricing
Prepaid credits, charged per task. See https://xzar.si/docs.html. Pilots and ongoing testing: https://xzar.si.
Frequently asked questions
What is Xzar?
Crash-test your AI agent on real professional decisions, graded against what actually happened.
How do I install Xzar?
It runs as a remote server. In Claude Code: claude mcp add --transport http xzar 'https://api.xzar.si/mcp'. For other clients, paste the JSON configuration from this page into their MCP settings.
Does Xzar need an API key?
Yes. It expects XZAR_API_KEY, which you set in its configuration.
Is Xzar open source?
Yes. Its code is on GitHub under the MIT license.
Who maintains Xzar?
It is published by xzar-si. AI.tn lists it but does not maintain or endorse it.
From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.