Does Codex Cloud support MCP servers, or can I copy config.toml as a workaround?
I'm using Codex Cloud (the web version, not the CLI) and want to use MCP servers such as Playwright MCP from the cloud. The CLI supports MCP, but Codex Cloud doesn't seem to expose MCP configuration. Could I work around this by copying my local ~/.codex/config.toml into the container's setup script so the configuration is applied inside the container? This issue is still open as of September 25, 2025.
Originally asked on GitHub: https://github.com/openai/codex/issues/4226
2 answers
Status: open — as of September 25, 2025, MCP server support in Codex Cloud is an open feature request, and the thread contains no maintainer response or confirmed workaround.
What's confirmed in the thread: MCP is supported in the Codex CLI, but not in Codex Cloud (the web/container environment). The issue author wanted to use Playwright MCP from the cloud and proposed copying their local ~/.codex/config.toml into the container's setup script so the configuration would be applied inside the container. That approach was never tested or confirmed by anyone in the thread, so treat it as a hypothesis.
If you need MCP today, the only option stated in the thread is to use the Codex CLI, which already supports MCP servers. For Codex Cloud, you'll have to wait for the feature and follow the issue for updates.
General advice (not from the thread): before trying the config.toml workaround, check whether Codex Cloud reads that file at all — if the web environment doesn't load ~/.codex/config.toml, copying it into the container won't change anything.
Security note: ~/.codex/config.toml can contain credentials such as API keys or tokens. Copying it into a cloud container exposes those secrets to that environment, so only do this if you know exactly what is in the file.
Did this work for you?What @jjangga0214 tried on GitHub (Outcome unclear)
The issue author proposed copying the local ~/.codex/config.toml into the container's setup script so MCP configuration would be applied inside the Codex Cloud container. This was only a question, not an actual test, and nobody in the thread confirmed whether it works.
Source: https://github.com/openai/codex/issues/4226 (opens external site)
Did this work for you?