Sync my chats, conversation history between ChatGPT website, Codex in VScode.

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #4825
• #5152
• #4268

Powered by Codex Action

This is related to #2153

I too would like to see more "unification" of things.

I really want a "superapp" that just has everything, on both macOS and iOS (or Windows/Android, etc.)

This would lead to less fragmentation, more shared context, and seamless workflows.

So have you come to any solution yet or still into that janky workaround??

It is impressive that this isn't native yet. Creating your own fragmentation is a choice.

I strongly support this request, but I’d like to add one important extension: true two-way sync between a ChatGPT conversation and a Codex task, not just shared chat history.

My normal workflow is:

ChatGPT planning / architecture → Codex implementation → ChatGPT review → Codex revision

Today, that loop requires a lot of manual context transfer. The ideal workflow would be:

1. From a ChatGPT thread, choose Send to Codex / Implement with Codex.
2. Codex receives the relevant approved plan, architecture, requirements, attachments, and project context from that thread.
3. The ChatGPT thread and Codex task become persistently linked.
4. While Codex is working, meaningful progress/status updates flow back into the originating ChatGPT thread (what it is implementing, files being changed, tests, blockers, important decisions).
5. When Codex finishes, the same ChatGPT thread automatically receives the implementation summary, changed files/diffs, test results, unresolved issues, commit info, and PR link when applicable.
6. ChatGPT can review those results using the original planning context, and follow-up instructions can be sent directly back to the same Codex task.
7. That loop can continue without repeatedly copying context back and forth.

So the model I’m asking for is essentially:

ChatGPT plan → Codex execute → Codex progress/results → ChatGPT review → Codex revise → repeat

For larger projects, ChatGPT works extremely well as the planning, architecture, product-design, and review layer, while Codex is the execution layer. Connecting those two layers with persistent two-way project context would remove a major source of friction.

A shared Project Context or linked ChatGPT/Codex session could be one way to implement it, while still keeping discussion and code execution clearly separated.

OpenAI Support specifically directed me to this GitHub tracker for Codex feature requests, so I’m adding this here rather than relying on the support thread alone.

I’d also like to request first-class synchronization and remote access specifically between the Codex desktop apps on macOS and Windows.

Desired workflow:

• Sign in with the same OpenAI account on a Mac and a Windows PC.
• See the same Codex projects, threads, chat history, titles, pins, and relevant task state on both computers.
• Continue a thread on either computer without manually copying ~/.codex/sessions or exporting summaries.
• Securely connect to and control Codex running on the other computer in either direction (Mac → Windows and Windows → Mac), including reviewing progress, answering questions, approving actions, and steering active tasks.
• Clearly distinguish cloud-synced metadata/history from machine-local files, credentials, permissions, terminals, and execution state.

For safety and reliability, an optional sync model with encryption, device management/revocation, explicit local-only threads, conflict handling, and clear offline behavior would be valuable. Synchronizing the live .codex directory through OneDrive/iCloud is unsafe because SQLite/session files can conflict, so an official solution would address an important cross-device workflow gap.

For anyone whose immediate need is read-only access to existing ChatGPT conversations from Codex, there is a useful unofficial workaround available now: shoyu-ramen/codex-chats-mcp.

It exposes ChatGPT conversations through a local STDIO MCP server, including list, search, and get operations. This is not a replacement for the broader native feature requested here: it does not synchronize local ~/.codex/sessions across machines, provide live two-way ChatGPT ↔ Codex task sync, or unify execution state. It is mainly useful for finding and reading prior ChatGPT conversations from Codex.

One important installation caveat as of codex-chats-mcp 0.1.1: its dependency is currently declared as mcp[cli]>=1.0, so a fresh install can resolve MCP Python SDK 2.0.0. The server still imports the v1-only mcp.server.fastmcp API and will then exit during startup with:

ModuleNotFoundError: No module named 'mcp.server.fastmcp'

Until the package pins v1 or migrates to the MCP SDK 2.x APIs, the verified workaround is to pin MCP SDK 1.29.0 inside the package's dedicated virtual environment:

/path/to/venv/bin/python -m pip install "mcp[cli]==1.29.0"
/path/to/venv/bin/python -m pip check

With Python 3.13.9, I verified that this restores the FastMCP import and allows an independent MCP initialize() and list_tools() handshake to complete successfully. Details and the upstream fix/migration proposal are tracked here: https://github.com/shoyu-ramen/codex-chats-mcp/issues/1#issue-5223152107

The server also offers rename/archive/delete operations, so anyone wanting a read-only setup should restrict the Codex MCP entry explicitly:

enabled_tools = [
    "list_conversations",
    "search_conversations",
    "get_conversation",
]