For AI-Assisted TeamsUse case

Your agent writes the code. Certyn checks the running app.

Connect Certyn to Claude, Cursor or your own client over MCP. The agent can run a real browser session against your staging app, read back what failed with the errors and a recording, and fix it before it tells you it is done.

Sounds familiar

Most of what shipped this week was written by an agent and skimmed once.
Unit tests pass and the page still renders blank.
The agent says the feature is done, and nobody opened the app to check.
You want the check inside the loop, not two days later in staging.

Start Free MCP setup2-minute setup

How it works

Write, run, look, fix. Without you in the middle.

Code review catches what the code says. Only running the app catches what it does. Giving the agent a browser closes that gap inside the same session.

Connect01
Connect Certyn to your client
One server URL and an OAuth sign-in with your normal Certyn account. Claude, ChatGPT and Cursor have pre-registered clients, and any other MCP client can register itself.
https://api.certyn.io/api/mcp
The agent02
The agent finishes a change
Instead of declaring victory, it calls a run tool: a quick test of the flow it touched, one of your existing test cases, or a whole process against your staging environment.
run_quick_test · run_test_case · run_process
The browser03
A real browser does the checking
Certyn signs in with the environment test account, walks the flow, and watches what the app actually renders and requests. This is not a mock and not a prediction.
What comes back04
The failure comes back as something to act on
The agent gets the failing step, console errors and failed requests as text it can use, plus a link to the recording for you. It can then fix the code and run again, or write the test case for what it just built.

The Certyn connections page listing MCP and app integrations available to connect — Everything an agent can be wired into, in one place: MCP servers for reading specs and tasks, trackers to file into, and the clients that can call Certyn back.

Try it

Give your agent a browser in two minutes.

Cursor reads this from your user or project mcp.json. Claude and ChatGPT take the same server URL as a custom connector.

Before you start

01Create a Certyn account and a project pointing at your staging URL.
02Add the test account to the environment as variables, so runs can sign in.
03Drop this into mcp.json and start the connection.
04Sign in when the client opens the Certyn login. No client secret and no API key.

Client IDs, scopes and the full tool list→

mcp.json

json

{
  "mcpServers": {
    "certyn": {
      "url": "https://api.certyn.io/api/mcp",
      "auth": {
        "CLIENT_ID": "tpc_jZgoggfpgCQ6q2ZwHeEr2B",
        "scopes": [
          "openid", "profile", "email", "offline_access", "mcp"
        ]
      }
    }
  }
}

Then try: "list my Certyn projects, run a quick test of the signup flow on staging, and tell me what failed". The Claude and ChatGPT connector IDs are in the setup guide.

In the app

What the agent finds, written down where you can check it.

Whatever the agent runs over MCP lands in the same project as everything else, with the reasoning attached to each finding.

A Certyn finding expanded to show the analysis, with actions to fix it with a coding agent — A finding with the expected and actual behavior spelled out, and a one-click hand-off to a coding agent to fix it.

What you get

What comes back to the agent

Structured enough for a model to act on, and detailed enough for you to check the model.

Attached to every runEvidence

A result, not a guess

Passed, failed or blocked for each step, from a session that really ran.

The failing step and the errors

Console errors and failed requests, in text the agent can reason about.

A recording for the human

When you want to see it yourself instead of reading a summary.

Test cases it can write

The agent can save the flow it just built as a case, so it runs from then on.

Project context

Wiki, existing cases and open tickets, so it knows what is already known.

Your permissions, unchanged

The connection sees exactly what your Certyn account sees, and nothing else.

Straight answers

Fair questions about putting this in the loop

Why not just have the agent write Playwright tests?

You can, and then you own that code. Selectors, fixtures, waits and a suite that the agent keeps growing and nobody prunes. Certyn runs flows from a description instead, so the agent checks its work without adding a second codebase to maintain. If you already have Playwright, keep it. This covers the part it does not: whether the app works for a person right now.

What can the agent actually change?

The connection is scoped to your own Certyn permissions, and nothing more. It can read projects, cases and tickets, run tests, and create or update test cases and wiki notes after you confirm. It has no access to your repository and cannot deploy anything.

Which clients work?

Claude, ChatGPT and Cursor have pre-registered OAuth clients you can paste in. Any other MCP client that supports dynamic registration can connect on its own. Sign-in is your normal Certyn login: MCP does not take API keys.

Is this slow enough to break the flow?

A quick test of one flow takes minutes, not hours, and the agent can keep working while it runs. It is far cheaper than the loop where a person finds the bug in staging two days later.

Can the same checks run in CI?

Yes, and they should. The cases the agent creates are ordinary Certyn cases, so the same ones run from GitHub Actions or any CI system on every pull request. The agent loop catches it early, CI keeps it caught.

What does a run cost?

Runs are billed as QA hours, and the free tier gives you one hour without a card. A quick test of a single flow is a few minutes of it, so you can try the whole loop before paying anything.

Wire it up

Guide→

Connect via MCP

Server URL, pre-registered client IDs for Claude, ChatGPT and Cursor.

Reference→

MCP tool catalog

Every tool, its arguments and what it returns.

Guide→

Ask Certyn skills

What the assistant can do with your project context.

Blog→

Exploratory testing with AI

What an agent finds when it walks an app with no script.

Get startedFree

Let the agent check its own work.

Start free, no card. Connect the MCP server and ask it to run your signup flow.

Start Free→Read the MCP guide

{ "mcpServers": { "certyn": { "url": "https://api.certyn.io/api/mcp", "auth": { "CLIENT_ID": "tpc_jZgoggfpgCQ6q2ZwHeEr2B", "scopes": [ "openid", "profile", "email", "offline_access", "mcp" ] } } } }