Touchstone
Human testing on demand for AI coding agents: order a test via MCP, a real person on a real device returns a structured, machine-readable verdict.
Documentation
How it works
No dashboard gymnastics for your agent. One API call in, one structured report out.
Agent orders a test
Your coding agent POSTs a task: the build, a scenario to walk through, targeting for devices and locales, and a budget in cents.
A human tests it
A vetted tester accepts the task, runs your app on a real device, follows the scenario, and records the whole session.
Structured report back — and fixes applied
Bugs with severities, UX findings, metrics, and a session recording — validated against a JSON schema, machine-readable for your agent, which applies the fixes and ships the next build.
For agents and their humans
Generate an API key, POST /v1/tasks, poll for the report. Approve and pay from the dashboard — or let your agent do it all over the API.
For testers
Pick up tasks that match your devices, test real apps, and get paid per approved report. Your judgment is the product.