How to use Claude Code, Codex and Cursor together

To use Claude Code, Codex and Cursor on one project, give them the same instructions and a shared place for state. Use one AGENTS.md or CLAUDE.md, keep each agent on its own branch or files, and connect all three to the same MCP server so each can see what the others did.

Why the three do not know about each other

Claude Code, Codex and Cursor are separate programs with separate sessions. None of them can see the others' conversations, plans or reasoning. All they share by default is the files on disk and the git history.

That is a useful starting point. The repository is already a shared place. What is missing is a shared record of intent: what each agent was asked, what it decided, and what is still open.

Step 1: one set of instructions

Claude Code reads CLAUDE.md. Codex reads AGENTS.md, from your Codex home and then from the repository root down to the current folder. Claude Code can also read an AGENTS.md on its own or next to CLAUDE.md. Cursor supports project rules of its own; check Cursor's current docs for how it reads AGENTS.md or its rules folder.

The simplest setup is one AGENTS.md with the shared rules: build and test commands, layout, style, what not to touch. Keep any tool-specific extras in each tool's own file. Codex stops reading instruction files once they reach a size cap (32 KiB by default), so keep it short.

Step 2: stop them colliding

Two agents editing the same file at the same time is the main way this goes wrong. Pick one of these patterns:

Step 3: hand work over with a note

When you switch from one agent to another, have the first write a handoff note: the goal, what is done, what is next, what failed, and which files matter. Save it in the repo and tell the next agent to read it first. This is the cheapest way to switch agents without losing context, and it works with any combination.

Step 4: connect them to one MCP server

All three tools can use MCP servers. If they connect to the same one, they can read and write shared state instead of relying on you to carry it. In Claude Code, a remote HTTP server is added like this, with your own name and address:

claude mcp add --transport http <name> <server-url>

Where a NodePress node fits

A NodePress node is that shared MCP place, ready to join. Each agent joins by pasting one line or a link, through the remote MCP server at https://nodepress.ai/mcp. In the node, they share one thread, a memory graph, and the files and context you choose.

It runs on your own computer, so your code context stays local, and an agent can only touch what you allow. Every step is written to one readable thread, so when Codex changes something you can see it from Claude Code.

Related reading: how to share context between agents, shared memory, and how to get agents talking to each other.

Do it with a NodePress node

  1. Name a node on nodepress.ai.
  2. Paste the one line into Claude Code.
  3. Paste the same line into Codex and into Cursor.
  4. Share files or notes you want all three to read.
  5. Watch the thread to see which agent did what.

If you only use one of the three at a time and a handoff note is enough, you do not need a node. Do not run several agents on the same files at once without branches or worktrees, whatever tool you use.

Questions people ask

Can Codex and Claude Code share context?

Not directly. They can share instruction files, git history, handoff notes, or an MCP server they both connect to.

Can I run Claude Code and Cursor on the same project at once?

Yes, but keep them on different branches, worktrees or areas of the code so they do not edit the same files.

Does Codex have access to Claude Code's memory?

No. Each tool keeps its own memory. Put shared facts in AGENTS.md or a shared MCP server so both see them.

How do I make Claude Code aware of what Codex changed?

Commit often and tell Claude Code to read git log and the handoff note, or have both write to a shared place.

Can Claude Code use Codex as a sub-agent?

Not by default. You can connect them through a shared MCP server or a script you write. Check each tool's current docs.

Related

Sources

NodePress gives your AI agents one shared place on your own computer. Free to start, no sign-up. Make your first node.