Overview

Playwright is an open-source, modern browser automation framework developed by Microsoft for testing and automating web applications. It enables developers and QA engineers to write reliable end-to-end (E2E) tests that simulate real user interactions across multiple browsers using a single API.

Unlike traditional automation frameworks, Playwright is designed to handle today's dynamic web applications, such as those built with React, Angular, Vue, and other JavaScript frameworks. It provides intelligent waiting mechanisms, robust element locators, built-in debugging tools, and cross-browser compatibility, making automated testing faster, more reliable, and easier to maintain. Playwright supports testing on Chromium (Google Chrome and Microsoft Edge), Mozilla Firefox, and WebKit (Safari), ensuring that applications behave consistently across different browsers and operating systems.

IBM Db2 Genius Hub (Db2GH) continues to evolve with agentic capabilities that transform database management into an AI-powered experience. As teams scale their Db2GH deployments, ensuring the reliability of its REST endpoints — particularly the QRep REST Central — through automated testing becomes critical.

This blog demonstrates how to use the Playwright Model Context Protocol (MCP) server together with IBM Bob (IBM's AI-powered developer assistant) to automate browser-based and API-level testing of Db2GH's QRep REST Central. We also provide a detailed comparison between Selenium WebDriver (legacy) and Playwright MCP (modern).

What is Playwright MCP and IBM Bob?

Playwright MCP server

The Playwright MCP (Model Context Protocol) server exposes Playwright's browser automation capabilities as a set of structured tools that any MCP-compatible AI client can call. Instead of writing test scripts manually, an AI agent can navigate pages, click elements, fill forms, capture screenshots, and assert page state — all through natural-language instructions that get translated into Playwright actions at runtime.

When Bob receives a test instruction, it internally coordinates three specialised agents to complete the work end to end:

These three agents work in sequence for every test scenario, making the entire test lifecycle — planning, execution, and self-healing — autonomous.

Key Playwright MCP tools

IBM Bob

IBM Bob is an AI-powered developer assistant embedded in the IBM toolchain. Bob uses MCP servers — including the Playwright MCP server — as tools. Bob can be instructed in plain English to "test the Db2GH login page", and it will coordinate the Planner, Test Generator, and Healer agents to execute the full test and interpret the results.

Architecture: How they work together

The integration follows a three-layer model:

Bob orchestrates all test steps — login, navigation, data entry, assertion — without any pre-written test script. The entire test specification lives in Bob's context as a prompt.

Selenium vs. Playwright — Feature comparison

The table below compares Selenium WebDriver and Playwright MCP feature by feature across every dimension relevant to modern test automation.

Code comparison: Login to Db2GH for Q replication

The same login action expressed in both frameworks:

Selenium (Java) — legacy

// Selenium WebDriver — Db2GH login
WebDriver driver = new ChromeDriver();
driver.get("https://db2gh.example.com/login");

WebDriverWait wait = new WebDriverWait(driver, Duration.ofSeconds(10));

WebElement username = wait.until(
    ExpectedConditions.visibilityOfElementLocated(By.id("username"))
);
username.sendKeys("admin1");

WebElement password = driver.findElement(By.id("password"));
password.sendKeys("admin1_password");

driver.findElement(By.cssSelector("button[type='submit']")).click();

wait.until(ExpectedConditions.titleContains("Db2 Genius Hub"));
System.out.println("Login successful: " + driver.getTitle());
driver.quit();

Playwright MCP + IBM Bob — modern

Bob receives a natural-language instruction and issues Playwright MCP tool calls:

// Bob prompt:
// "Log into Db2 Genius Hub at https://db2gh.example.com with username admin1
//  and password admin1_password. Confirm the dashboard loads."

// Bob issues these Playwright MCP calls automatically:
browser_navigate({ url: "https://db2gh.example.com/login" })
browser_snapshot()   // ← Bob reads the page structure
browser_fill_form([
  { target: "#username", name: "Username", type: "textbox", value: "admin1" },
  { target: "#password", name: "Password", type: "textbox", value: "admin1_password" }
])
browser_click({ target: "button[type='submit']", element: "Submit" })
browser_wait_for({ text: "Dashboard" })
browser_take_screenshot({ type: "png", scale: "css" })

Setting up Playwright MCP with IBM Bob

Prerequisites

• IBM Bob installed and licensed (version 1.5 or later)
• Node.js 18+ installed on the workstation
• Access to a running Db2 Genius Hub for testing purposes

Step 1: Install the Playwright MCP server

npm install -g @playwright/mcp@latest
npx playwright install chromium

Step 2: Register the MCP server in IBM Bob

Add the following entry to Bob's MCP server configuration file (~/.bob/mcp_servers.json):

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp", "--headless"],
      "env": {}
    }
  }
}

For headed mode (visible browser window, useful for debugging):

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp"],
      "env": {}
    }
  }
}

Alternatively, MCPs can be configured from the settings panel in IBM Bob.

Step 3: Verify the connection

Ask Bob to run a quick health check:

// Bob prompt:
"Navigate to https://db2gh.xxx.com and take a screenshot."

// Expected: Bob responds with a screenshot of the Db2GH login page.

Test example

Example: Login to Db2 Genius Hub (platform) for Q replication

This example walks through a full login test for Db2GH and shows exactly how the three Playwright MCP agents — Planner, Test Generator, and Healer — each play their part.

Bob prompt:
"Test the login page of Db2 Genius Hub at https://db2gh.example.com/login. Log in as Aadmin11234 / Aadmin1_password1234 and confirm the dashboard loads."

Step 1 — Planner agent: Analyse the page

The Planner agent opens the URL, captures the live accessibility tree, and produces a structured test plan before any test code is written.

// Planner agent — understand the page structure
browser_navigate({ url: "https://db2gh.example.com/login" })
browser_snapshot()
// ↳ Planner reads the accessibility tree and identifies:
//   - role="textbox" labelled "Username"  → selector: #username
//   - role="textbox" labelled "Password"  → selector: #password
//   - role="button"  name "Sign in"       → selector: button[type='submit']
//   - Success indicator: text "Dashboard" present after login

Planner output — test plan:

1. Navigate to https://db2gh.example.com/login
2. Fill Username field with "admin1"
3. Fill Password field with "admin1_password"
4. Click the Sign in button
5. Assert the text "Dashboard" is visible on the resulting page
6. Capture a screenshot as evidence

Step 2 — Test generator agent: Write and execute the test

The Test Generator agent translates each plan step into Playwright MCP tool calls and executes them against the live browser in sequence.

// Test generator agent — execute the plan
browser_fill_form([
  { target: "#username", name: "Username", type: "textbox", value: "admin1" },
  { target: "#password", name: "Password", type: "textbox", value: "admin1_password" }
])
browser_click({ target: "button[type='submit']", element: "Sign in" })
browser_wait_for({ text: "Dashboard" })
browser_take_screenshot({ type: "png", scale: "css",
                          filename: "login-success.png" })
// ✓ Dashboard text found — test passes

The agent will also write all possible test scenarios from the login page.

Step 3 — Healer agent: Detect and correct failures

Suppose a Db2GH UI update changed the submit button selector from button[type='submit'] to button[data-action='login']. The generated test would fail. The Healer agent intercepts the failure and fixes it automatically.

// Healer agent — triggered on step failure

// ✗ browser_click({ target: "button[type='submit']" }) → Element not found

// Healer re-reads the live accessibility tree
browser_snapshot()
// ↳ Healer finds: role="button" name="Sign in" → new selector: button[data-action='login']

// Healer retries with the corrected selector
browser_click({ target: "button[data-action='login']", element: "Sign in" })
browser_wait_for({ text: "Dashboard" })
browser_take_screenshot({ type: "png", scale: "css",
                          filename: "login-success-healed.png" })
// ✓ Step healed and test passes — no human intervention required

Healer report Bob surfaces:

⚠ Step 4 failed: selector "button[type='submit']" not found.
✔ Healer re-derived selector from accessibility tree: "button[data-action='login']"
✔ Step retried successfully. Test passed.
    Screenshot saved: login-success-healed.png

Summary

In this blog, we covered:

• The architecture of Playwright MCP + IBM Bob as an AI-driven test automation platform for Db2 Genius Hub Qreplication.
• The three-agent model — Planner, Test Generator, and Healer — that makes test authoring and self-healing autonomous
• A comprehensive feature comparison between Selenium WebDriver and Playwright MCP across 27 dimensions
• A concrete test example showing all three agents in action

By adopting Playwright MCP with IBM Bob,  teams eliminate the overhead of maintaining Selenium test infrastructure while gaining AI-assisted test authoring, built-in REST API assertion, and tighter integration with the IBM developer toolchain.