Skip to main content

​
Overview

OneTest uses an AI-powered test execution engine through the OneTest QA Agent. Tests stored in OneTest are executed via the /qa-onetest skill in Claude Code, which drives a real browser, records results per step, and reports everything back to OneTest automatically.

​
Prerequisites

1

Install Claude Code

Install Claude Code — the CLI tool that runs the QA agent.
2

Install the QA Agent

Clone the QA agent skills into your project:
git clone https://github.com/onetest-ai/qa-agent.git
3

Configure the OneTest MCP Server

Tip: Run npx github:onetest-ai/qa-agent init and choose “Connect via browser” for one-click setup — it configures the MCP connection automatically.
Or add the test-management MCP server manually to your .mcp.json (project root):
{
  "mcpServers": {
    "test-management": {
      "type": "http",
      "url": "https://tms.onetest.ai/mcp/test-management",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}
Get your API key from tms.onetest.ai -> Settings -> API Keys.
4

Restart Claude Code

Restart Claude Code so it picks up the new MCP server configuration.

​
Running a Test Run

Use the /qa-onetest run command in Claude Code to create, execute, and complete a full test run. 
/qa-onetest run
The agent will prompt you for which tests to include, then automatically:
1

Create Run

Creates a test run in OneTest with the selected test cases.
2

Start Run

Transitions the run from planned to in_progress and freezes the scope.
3

Execute Tests

For each test case in the run:
  • Navigates to the target URL in a real browser
  • Executes each test step via Chrome DevTools Protocol (CDP)
  • Captures screenshots before and after each step
  • Checks the browser console for errors after each step
  • Validates actual results against expected outcomes
4

Record Results

Records the result for each test execution:
  • Pass / Fail / Blocked / Skipped status
  • Step-level results with screenshots
  • Failure reasons and classification
  • Links to defects if applicable
5

Complete Run

Marks the run as complete and generates analytics — pass/fail rates, completion percentage, duration, and failure breakdown.

​
All Commands

The /qa-onetest skill supports four commands:

run

Create, execute, and complete a full test run. The agent drives a real browser and records all results back to OneTest.
/qa-onetest run

push findings

Convert QA audit findings into OneTest test cases. Maps priorities (p0-p3) and categories (accessibility, security, performance, etc.) automatically.
/qa-onetest push findings

pull tests

Fetch test cases from OneTest for local browser execution.
/qa-onetest pull tests

status

Show the execution queue and any active runs.
/qa-onetest status

​
How Execution Works

​
Browser-Based Testing

The agent executes tests in a real Chrome browser using the Chrome DevTools Protocol (CDP). This means tests interact with your application exactly as a user would — clicking buttons, filling forms, navigating pages, and verifying visual output.

​
Exploratory Findings

If the agent discovers issues during test execution that aren’t covered by existing test cases, it can record them as exploratory findings using record_exploratory_result. These appear in your test run alongside the planned test results.

​
Parallel Execution

Non-conflicting tests can run simultaneously using separate Chrome instances on different CDP ports. This speeds up execution when tests target different URLs or isolated user sessions.

​
Recording Results

Every test execution records detailed results back to OneTest:
Each step in a test case gets its own pass/fail status, along with:
  • Screenshots captured before and after the step
  • Console output from the browser
  • Actual vs. expected result comparison
When a test fails, the agent records:
  • Failure reason: What went wrong
  • Failure classification: Environment issue, product defect, test data issue, etc.
  • Defect links: Links to related bug tracker issues
After the run completes, OneTest provides analytics:
  • Pass/fail/skip/blocked rates
  • Completion percentage
  • Total execution duration
  • Failure breakdown by category

​
Typical Workflow

A common end-to-end workflow looks like this:
  1. Audit your site with QA specialist skills (/qa-accessibility, /qa-security, etc.)
  2. Push findings to OneTest as test cases with /qa-onetest push findings
  3. Run tests with /qa-onetest run to execute them in a real browser
  4. View results in the OneTest dashboard to analyze trends and track regressions

​
What’s Next?

Viewing Results

Analyze test results and trends in the dashboard

QA Agent on GitHub

Full documentation for the QA agent and all skills