Playwright MCP — Parallel Instances (Unofficial Fork)

Disclaimer: This is an unofficial, modified fork of Microsoft Playwright. It is not affiliated with, endorsed by, or maintained by Microsoft. Use at your own risk.

Why This Fork Exists

The official Playwright MCP server runs a single shared browser context — all tool calls share one set of tabs, cookies, and state. This makes it impossible for multiple AI agents (or sub-agents) to browse the web in parallel without stepping on each other.

This fork is a workaround that adds browser instance isolation so each agent gets its own independent session through the same MCP server. It's a pragmatic hack, not a polished product — if/when upstream Playwright adds native parallel support, this repo becomes obsolete.

What's Changed

Three new tools and one new parameter on all existing tools:

When instanceId is omitted, tools operate on the default instance (fully backward-compatible with standard Playwright MCP).

Quick Start

Prerequisites

• Node.js ≥ 18
• Chromium/Chrome (auto-installed by Playwright if needed)

Build from Source

git clone https://github.com/martin-lzh/playwright-mcp-parallellel.git
cd playwright-mcp-parallellel
npm install
npm run build

Install Browsers (first time only)

npx playwright install chromium

Start the MCP Server

HTTP/SSE (for networked agents):

node packages/playwright-core/lib/tools/mcp/cli-stub.js --isolated --port 3000

Stdio (for local MCP clients like VS Code):

node packages/playwright-core/lib/tools/mcp/cli-stub.js --isolated

Shortcut:

npm start
# → node packages/playwright-core/lib/tools/mcp/cli-stub.js --isolated --port 3000

VS Code / Copilot Configuration

Add to your VS Code settings.json:

{
  "mcp": {
    "servers": {
      "playwright": {
        "command": "node",
        "args": [
          "C:/absolute/path/to/playwright-mcp-parallel/packages/playwright-core/lib/tools/mcp/cli-stub.js",
          "--isolated"
        ]
      }
    }
  }
}

Replace the path with your actual clone location.

Usage — Parallel Agents

Each agent creates its own instance, uses it, then cleans up:

Agent A                              Agent B
───────                              ───────
browser_instance_create              browser_instance_create
  → instanceId: "agent-a"             → instanceId: "agent-b"

browser_navigate                     browser_navigate
  url: "https://site-a.com"           url: "https://site-b.com"
  instanceId: "agent-a"               instanceId: "agent-b"

browser_snapshot                     browser_click
  instanceId: "agent-a"               element: "Submit"
                                       instanceId: "agent-b"

browser_instance_close               browser_instance_close
  instanceId: "agent-a"               instanceId: "agent-b"

Tools called without instanceId hit the default instance — works exactly like upstream.

CLI Flags

Development

npm run build                       # Full build
npm run ctest-mcp                   # Run all MCP tests (Chromium)
npm run ctest-mcp -- instance       # Instance isolation tests only

License & Attribution

This project is a modified fork of Microsoft Playwright, which is licensed under the Apache License 2.0.

The original copyright and license terms are preserved in full — see LICENSE. Per the Apache 2.0 license terms, this fork constitutes a derivative work. The modifications add browser instance isolation to the MCP server; all original Playwright functionality remains intact and is the work of Microsoft and the Playwright contributors.

This project is not an official Microsoft product.