Feedback from a remote Codex helper prototype: long-session performance, durable queue, and MCP process reuse

Hi everyone,

I wanted to share some practical Codex feedback and a small prototype direction I have been working on.

I use Codex heavily every day. Recently I built a personal web interface that lets me work remotely with Codex running on my home PC from a laptop or phone. Shortly after that, OpenAI shipped a similar remote-work direction, which was exciting to see.

While building and using my prototype, I ran into a few issues that may be useful for the Codex team and other heavy users.

1. Very large session history files

Long-running Codex conversations can produce rollout/history JSONL files that grow to hundreds of MB. I have several sessions around 500 MB or larger.

With those sessions, the native Codex app can freeze for a noticeable amount of time, especially when sending a new prompt.

In my prototype, I worked around this by indexing session summaries and recent user/assistant messages into local SQLite. The UI reads summaries and recent messages from SQLite, reads only bounded head/tail windows for active sessions, and lazy-loads heavy technical records such as command output, tool output, and event records only when the user opens them.

This keeps the UI responsive without modifying Codex’s original JSONL files.

2. Durable queued and delayed messages

The Codex app already has a queue, but I made the queue durable by storing pending messages in SQLite. If the app/helper is closed or restarted, queued messages are not lost. When the current task finishes, saved messages are sent automatically.

I also implemented time-based delayed tasks using the same queue storage and processing model.

3. Duplicate MCP server processes

I also noticed that Codex may start duplicate MCP server instances across conversations, even when the MCP server command/configuration is the same.

On my home PC, I recently saw 14 instances each of four MCP servers, about 56 processes total. I also saw a similar pattern on my MacBook with repeated serena, context7, node_repl, and codex app-server --listen stdio:// processes.

Maybe this could be improved with a shared MCP runtime, reference-counted server lifecycle, or more aggressive cleanup of inactive MCP processes.

Links

I posted the broader GitHub Discussion here:

GitHub

Ideas from a remote Codex helper prototype: faster long-session history,...

Hi Codex team, First of all, thank you for the work on Codex. I use it heavily every day, and it has become one of my main development tools. Recently I built a personal web interface for working r...

And two focused GitHub issues:

• Large session history performance: Codex app can freeze with very large rollout/history JSONL files · Issue #22991 · openai/codex · GitHub
• Duplicate MCP server processes: Codex app appears to start duplicate MCP server processes across conversations · Issue #22992 · openai/codex · GitHub

I would be happy to share more implementation details or show a demo if this overlaps with anything the Codex team is thinking about.

Thanks again to everyone working on Codex. It has become a core part of my daily development workflow.