[Bug] Running multiple parallel async browsers in --isolated mode leaves orphan browsers. This has started happening after Release v0.0.69.

[ShishirPrashanth]

Bug Description

I created multiple concurrent HTTP sessions through a simple Python script using asyncio. Multiple browsers opened and worked as expected, and the returned results were correct.

However, after execution completed, only one browser terminated successfully. The remaining N-1 browser instances were left orphaned.

This happens after the completion of the last (slowest) task.

NOTE: This issue started happening from v0.0.69 onwards. It was not happening till v0.0.68.

---

Reproduction Steps

MCP Server Setup

npx @playwright/mcp@latest --port 8931 --isolated

---

Client Code

import asyncio

import httpx
from mcp import ClientSession
from mcp.client.streamable_http import streamable_http_client
from mcp.shared._httpx_utils import create_mcp_http_client

TARGET_URLS = [
    "https://www.wikipedia.org",
    "https://www.google.com",
    "https://www.amazon.in",
]

MCP_CODE = """
async (page) => {
    await page.waitForLoadState('networkidle');
    await page.waitForTimeout(2000);
    return await page.content();
}
"""


async def get_page_html(url: str) -> str:
    mcp_url = "http://localhost:8931/mcp"

    long_timeout = httpx.Timeout(300.0, read=300.0)

    async with create_mcp_http_client(timeout=long_timeout) as http_client:
        async with streamable_http_client(
            mcp_url,
            http_client=http_client,
        ) as (read, write, _get_session_id):

            async with ClientSession(read, write) as session:
                await session.initialize()

                await session.call_tool(
                    "browser_navigate",
                    {"url": url},
                )

                tool_result = await session.call_tool(
                    "browser_run_code_unsafe",
                    {"code": MCP_CODE},
                )

                return tool_result


async def main():
    tasks = [get_page_html(url) for url in TARGET_URLS]

    results = await asyncio.gather(*tasks)

    for url, result in zip(TARGET_URLS, results):
        print(f"URL: {url}")
        print(result)
        print("-" * 100)


if __name__ == "__main__":
    asyncio.run(main())

---

Observed Behavior

• 3 browser instances open successfully.
• All requests complete successfully.
• Correct results are returned.
• Only one browser instance closes at the end.
• The remaining 2 browser instances remain orphaned.

---

Expected Behavior

All browser instances created for concurrent sessions should terminate cleanly after task completion.