Guide
How to test the app your coding agent built
Reading the diff tells you what changed. Using the app tells you whether it works for someone new. Here is how to make the second part automatic, inside your coding agent's loop.
How should I test code an AI wrote?
Run your tests, then use the change the way a user would, on a preview deployment, before you review the diff. Mimiq automates the second part: your coding agent calls it through the MCP server, AI agents try the preview as different users in real browsers, and the result (what broke, failed or slow requests, where they got stuck) comes back to the coding agent so it can fix it and run again.
Green tests, broken app
Coding agents write code that passes the tests they also wrote. The unit tests are green and the signup still fails, because the form posts to a route that exists only locally, or the new button is hidden behind a modal on a phone. Nothing in the diff says so. Using the app does.
- Your coding agent calls test_flow on the preview it just built.
- AI agents try it as users and report what broke, with steps and screenshots.
- Browser records: JavaScript errors, failed and slow requests.
- The coding agent fixes and reruns before you review.
- Works with Claude Code, Cursor, Codex, VS Code and Windsurf.
How it works
- 01
Connect the MCP server
In Claude Code: claude mcp add --transport http mimiq https://mcp.mimiqai.com/mcp. Other clients take the same address. One free test works without a key.
- 02
Add the rule
Put the two-line rule below in CLAUDE.md or AGENTS.md, so the check runs on every user-facing change without you asking.
- 03
Let it loop
After a change, the coding agent deploys or finds the preview, runs the agents, reads what they hit, fixes it and runs them again.
- 04
Review with evidence
The summary links the report: you review the diff knowing new users got through.
The rule
A short, specific rule is easier for a coding agent to follow than a long overview:
## Before calling a user-facing change done
After changing anything a user goes through (signup, onboarding, checkout,
settings, a new feature), run mimiq.test_flow on the preview URL with the task
a user would try. Fix what the agents hit, then run it again. Ask me before
using more than 25 credits.Local builds
Agents browse from Mimiq's servers, so the address must be public. Use the preview deployment, or a temporary tunnel such as cloudflared for a local dev server. The page is public while the tunnel runs, so the coding agent should ask before it opens one.
Limits
Agents are simulated users. They find problems fast and on every build; they do not forecast how many real people will convert, they are not a deterministic regression suite, and they do not install native mobile apps.
Questions
Does my coding agent need a key?
Not for the first test: one free run of up to 3 agents without a key. A free account adds 50 credits and a key, and saves every run as a report.
How much does each check cost?
5 credits an agent, so a check with 3 agents uses 15. Then it is a $9 Starter (120 credits that never expire) or monthly plans from $19.
Will it slow my agent down?
A run of 3 agents takes about 5 minutes. Most coding agents keep working on other files while it runs, or you can run it before review instead of after every edit.
Keep reading
Let the agents find what breaks first.
Your first test is free: sign in with Google and it starts right away. Last checked 2026-10-22.