Chrome-Devtools-MCP Error on Linux

I don't think this is Codex related but if it works in another agentic coding tool and not in Codex, feel free to reopen.

Same thing happens to me as well, I can use chrome-devtools with copilot and claude code but not with codex

Same thing here - works in Gemini and Claude Code, but reports the aforementioned error in Codex CLI (Full Permissions granted).

Could it be related to sandboxing?
https://github.com/openai/codex/issues/1124 ?

But there are no more .ts code its written so it could be related to https://github.com/openai/codex/blob/774892c6d7ddbe55bf8862748f9507b03b75c92f/codex-rs/core/src/landlock.rs#L16

1. Configure the browser-url parameter for chrome-devtools mcp
2. Run the following command to open Chrome.

nohup google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-debug-headful --no-first-run --disable-gpu about:blank >/tmp/chrome-launch.log 2>&1 &
A workaround to bypass the codex sandbox, even when I use --yolo.

I was able to get the following successful result form the Chrome MCP using the prompt:

Please check the LCP of web.dev.

which was suggested by their docs. Could somebody provide more specific set of examples/cases of what isn't working.

The Chrome MCP server is what will execute its chrome commands, not our sandboxed exec tool so the sandbox shouldn't be in play here.

<img width="2986" height="2622" alt="Image" src="https://github.com/user-attachments/assets/ff8cde99-db6c-44d8-8a3c-bd45bce8a36f" />

same thing on ubuntu

same thing on ubuntu

Does it work with other coding agents?

Same thing on Linux Mint, works fine with Claude Code

LMDE7 is the same; both Codex and Claude Code have this issue.

Same thing here - works in Gemini and Claude Code, but reports the aforementioned error in Codex CLI (Full Permissions granted).

1. Configure the browser-url parameter for chrome-devtools mcp 2. Run the following command to open Chrome. nohup google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-debug-headful --no-first-run --disable-gpu about:blank >/tmp/chrome-launch.log 2>&1 & A workaround to bypass the codex sandbox, even when I use --yolo.

Thanks, It worked

@LittleTurtle2333 are you also on Linux?

@LittleTurtle2333 are you also on Linux?

Yes, I also use a Linux-based system.

> 1. Configure the browser-url parameter for chrome-devtools mcp > 2. Run the following command to open Chrome. > nohup google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-debug-headful --no-first-run --disable-gpu about:blank >/tmp/chrome-launch.log 2>&1 & > A workaround to bypass the codex sandbox, even when I use --yolo. Thanks, It worked

What parameter should be used for the browser-url?

Just gotta it:
"--browser-url=http://127.0.0.1:9222"

Using the following to set up the mcp seems to work

[mcp_servers.chrome-devtools]
command = "bash"
args = [
"-lc",
"DISPLAY=:1 XAUTHORITY=$HOME/.Xauthority npx -y chrome-devtools-mcp@latest --isolated"
]

[mcp_servers.chrome-devtools]
command = "npx"
args = ["chrome-devtools-mcp@latest", "--executablePath", "/usr/bin/chromium", "--headless", "--isolated"]
env = {}

This is what worked for me on opensuse. It seems to work fine with headless.

I have same issue on wayland ... seem like mcp exec in sandbox and no idea to lanuch chrome

Same issue on wayland. I'm using Manjaro. Please help. Thanks.

Same issue on wayland. Please help. Thanks.

i find a workaround to use this mcp with headful but need manually run chrome before

[mcp_servers.chrome-devtools]
args = [
    "-y",
    "chrome-devtools-mcp@latest",
    "--browser-url=http://127.0.0.1:9222"
]
command = "npx"
startup_timeout_sec = 60
# NOTE: run cmd before `google-chrome-stable --remote-debugging-port=9222 --user-data-dir=/tmp/chrome-profile-stable --no-first-run`

ref: https://github.com/ChromeDevTools/chrome-devtools-mcp?tab=readme-ov-file#manual-connection-using-port-forwarding

Edit: @BeardTech workaround is more solid! Check it out: https://github.com/openai/codex/issues/5166#issuecomment-3824698247

---

On Fedora with Wayland and XWayland available:

[mcp_servers.chrome-devtools]
command = "npx"
args = ["-y", "chrome-devtools-mcp@latest"]
env = { DISPLAY = ":0", GDK_BACKEND = "x11", QT_QPA_PLATFORM = "xcb", SDL_VIDEODRIVER = "x11" }

Then run xhost +SI:localuser:$(whoami) to allow local connections to XWayland.

Restart codex.

It worked for me.

Thanks to Codex, I found a working fix for Wayland.

✅ Fix

Edit ~/.codex/config.toml and pass Wayland env vars + force Chrome’s Wayland backend:

[mcp_servers.chrome-devtools]
command = "npx"
args = [
  "-y",
  "chrome-devtools-mcp@latest",
  "--chrome-arg=--ozone-platform=wayland",
  "--chrome-arg=--enable-features=UseOzonePlatform"
]
env_vars = ["WAYLAND_DISPLAY", "XDG_RUNTIME_DIR", "DBUS_SESSION_BUS_ADDRESS"]

✅ Result

Chrome launches headful under Wayland without switching to X11 or using Xvfb.

@BeardTech Wow! Amazing! It works.

This bug report hasn't received any new upvotes or posts in about two months, so I presume that it no longer occurs or the above workaround suffices.