Author Tests with Cursor & Claude Code

Connect Cursor, Claude Code or Claude Desktop to CertiQA through its MCP server so your AI assistant drafts test suites, and your team reviews them as plain steps in the visual canvas instead of reading code.

What This Gives Your Team

Writing a first test suite for a feature is the slow part. An AI coding assistant is good at that first draft: it reads your code, your tickets or your PRD and proposes the cases. The risk with AI-written tests has always been review – someone has to read the generated code to know whether the tests are any good. With CertiQA the assistant does not hand you code. It creates projects, suites, cases and steps directly in CertiQA through the MCP server, and they appear in the same visual step canvas your team already uses. A reviewer reads a list of plain-language steps and assertions, edits what is wrong, and runs the suite.

The workflow in one line

• Ask your assistant to draft tests -> it calls CertiQA tools -> the cases show up in your project -> you review the steps in the canvas -> you run the suite and gate CI on it.

1. Connect Your Assistant

There are two ways to connect. The remote server with OAuth sign-in is the easiest for individuals: no API key to copy, and you sign in as yourself. The local server with an API token suits shared setups and scripts.

Option A: Remote MCP server with OAuth (recommended)

In the CertiQA app open Settings -> API Keys and find the Connected AI Clients & Remote MCP card. It shows your remote MCP server URL and ready-made setup for each client. For Claude Desktop and Cursor, one command writes the configuration for you:

# Claude Desktop
npx -y @certiqa/mcp-server install --client claude --url <your-mcp-url-from-settings>

# Cursor (writes .cursor/mcp.json in your workspace)
npx -y @certiqa/mcp-server install --client cursor --url <your-mcp-url-from-settings>

Restart the client, then sign in to CertiQA when it prompts you. For Claude Code, add the same remote URL with its own command, then complete the sign-in when prompted:

# Claude Code
claude mcp add --transport sse certiqa <your-mcp-url-from-settings>

Option B: Local server with an API token

Create a scoped token under Settings -> API Tokens and add this to your client’s MCP configuration (Claude Desktop’s claude_desktop_config.json, Cursor’s .cursor/mcp.json, or the equivalent for your tool). Keep the token in a secret manager or environment variable, never in a file you commit.

{
  "mcpServers": {
    "certiqa": {
      "command": "npx",
      "args": ["-y", "@certiqa/mcp-server"],
      "env": {
        "CERTIQA_API_URL": "https://api.certiqa.app",
        "CERTIQA_API_TOKEN": "your_scoped_token"
      }
    }
  }
}

Any assistant that supports MCP can use these. The one-command installer covers Claude Desktop and Cursor; other clients, including Claude Code, use their own MCP add command or configuration file with the same URL or the same server command.

2. Ask for a Draft

Start with a specific request and name the project. The assistant has CertiQA’s tools, so it can look at what already exists before it writes anything, and it can reuse a CertiBlock (for example your login flow) instead of copying steps.

Using the certiqa tools, list my projects. In the "Storefront" project, create a suite called
"Checkout - guest" with cases for: successful card payment, declined card, expired promo code,
and an out-of-stock item at the last step. Reuse the existing login CertiBlock where a user
needs to be signed in. Use the recommended step format with several locator candidates per step.
Do not run anything yet.

For an AI agent under test the same approach works. Ask the assistant to create agent cases (conversation turns, mocked tool responses and expected tool calls) for the behaviors you care about.

What the assistant can do	Example tools
Find and organize work	list_projects, create_project, list_suites, create_suite, create_scenario
Create and edit tests	create_case, create_agent_case, update and delete cases, add cases to suites, reuse CertiBlocks
Run and schedule	trigger a run, schedule a recurring run, cancel a run
Investigate results	read run results, failures and artifacts, triage a failing case

The full list of 36 tools is in the MCP Server & Tools reference.

3. Review Without Reading Code

Open the project in CertiQA. The new cases are in the suite the assistant created, shown in the visual step canvas: each step is a plain action (open this page, type this value, click this button, expect this text) and each agent case shows its turns, mocked tool responses and assertions. Review them the way you would review a test plan:

A quick review checklist

• Does each case match a behavior you actually want protected, and is anything important missing?
• Are the assertions specific? “Page loads” is weak; “order total shows $149.00” is a real check.
• Do the steps use the right locators and test data? Fix anything the assistant guessed.
• For agent cases, are the mocked tool responses realistic and do the expectations cover the failure paths?
• In the recommended step format, each step records that it was written by an AI and how confident it was. CertiQA’s self-healing and review tooling use this, so ask the assistant to include it and check the lowest-confidence steps first.

Edit steps directly in the canvas, or ask the assistant to revise them. When you are happy, run the suite from the app or tell the assistant to trigger it.

4. Put It in CI

Once the suite is reviewed, gate your pipeline on it exactly as you would for hand-built tests. The CI runner script returns a pass or fail exit code. See the CI Runner guide for the setup.

Plans and Limits

Plan	AI assistant access
14-day trial	Connect Claude Desktop, Cursor and other MCP clients and author tests with your own assistant and LLM.
Starter Solo and Team Pro	Same interactive authoring with your own assistant and LLM.
Team Pro Plus and Enterprise Ultra	Everything above, plus the full MCP server suite for headless CI orchestration and cloud-hosted triage agents that use scoped agent tokens.

Your assistant uses your own Cursor or Claude subscription for the drafting. Running the tests it creates uses credits the same as any other run.

Safety and Good Habits

Keep a human in the loop

• An assistant can create, change and delete cases and trigger runs. Review what it created before you trust it, and keep important suites under your normal change process.
• Sign-in via OAuth acts as you, so the assistant has your permissions in CertiQA. Use a scoped token with limited permissions if you want a narrower grant.
• Never paste tokens or secrets into a prompt. Put them in the client’s environment configuration.
• Ask the assistant not to run anything until you have reviewed the draft, as in the example prompt above.

• 🛰️ AI & Agent Enablement — llms.txt, the example skill and the step format reference.
• 🤖 Composing Agent Tests Visually — The canvas where you review the steps the assistant creates.