Developers
Call the verdict from anywhere
ArbiterQA is headless and agent-native. Drive it from the arbiterqa npm CLI, an MCP client, or the public catalog API — name checks and sets by the same ids you see in the catalog and get a structured pass, fail, or error back.
npx arbiterqa
The CLI wraps the same API. Install nothing to try it, authenticate once, then run checks or sets in CI. Published on npm.
# Run without installing
npx arbiterqa --help
# Or add it to your project
npm install --save-dev arbiterqa# Authenticate the CLI (opens the browser, lands on /cli-success)
npx arbiterqa login# Run a single check against a URL
npx arbiterqa run \
--url https://example.com \
--validation font-size-body-min-max
# Run a whole validation set and fail CI on a failed verdict
npx arbiterqa run \
--url https://example.com \
--set accessibility-compliance \
--fail-on failModel Context Protocol
Point any MCP client at the hosted endpoint to give an agent the catalog and the ability to run checks. The endpoint lives on the API host: https://api.arbiterqa.com/mcp
{
"mcpServers": {
"arbiterqa": {
"url": "https://api.arbiterqa.com/mcp"
}
}
}Public read endpoints
The catalog, per-check detail, and sets are public and require no auth — the same JSON that powers this site and the MCP get_validation tool. Running a job (which spends credits) happens on the app behind your API key.
# Public, no-auth catalog reads
curl https://api.arbiterqa.com/api/validations
curl https://api.arbiterqa.com/api/validations/font-size-body-min-max
curl https://api.arbiterqa.com/api/validation-sets
curl https://api.arbiterqa.com/api/validation-sets/accessibility-complianceFront-door files
Agents can read our capabilities directly. These stay on the API host where agents expect them.
Get an API key
Create a free account with 1,000 credits, generate a key, and run your first check in minutes.