How to Migrate from Cursor to meshcode: A Step-by-Step Guide

You don't need to export anything special to leave Cursor — your code was never inside Cursor to begin with, it's just sitting in your repo like it always was. The actual work of switching is smaller than people expect: point meshcode at the same folder, connect the model subscription you already pay for, and adjust to a pane-based layout instead of a single window. This guide walks through that process step by step, plus what genuinely changes in your daily workflow and what doesn't.

Before you start: what "migrating" actually means here

There's no project file, no proprietary format, no database to export. Cursor is a VS Code fork reading your files off disk; meshcode is a native app doing the same thing. Migrating is really three separate, small tasks:

1. Open your existing repo in meshcode.
2. Connect your Claude and/or Codex CLI so you're not starting from zero on model access.
3. Rearrange your mental model from "one agent, one window" to "one agent per pane."

None of this touches your git history, your .env files, your CI, or your deploy pipeline. If it lives in the repo, it moves with the repo — because it never left.

Step 1: Open your project (no import, no conversion)

• Install meshcode (Mac or Windows) and launch it.
• Open your existing project folder — the same one Cursor was pointed at, .git and all.
• meshcode reads your working tree directly. There's no proprietary index to rebuild and nothing to wait on before you can start prompting.

If you had a .cursorrules file or similar project instructions, keep it — plain-language project notes in a text file are still useful context for any agent, meshcode included. You're not required to rewrite anything to get moving.

Step 2: Bring your Claude or Codex subscription with you

This is the step most people ask about, because it's the one that saves money. If you already pay for Claude or ChatGPT, you don't need a second AI subscription to use meshcode:

• In meshcode, connect your Claude or Codex account through their own CLI.
• Once connected, that pane runs on your existing subscription — with no extra token charge from meshcode.
• If you'd rather not connect anything yet, skip this step and use the built-in meshcode model stack instead, which runs on pay-as-you-go credit.

Most people end up doing both: their own Claude in one pane for the work they'd normally do in Cursor, and the built-in meshcode model in another pane for the cheaper, higher-volume tasks — boilerplate, mechanical refactors, test scaffolding — so the expensive subscription is reserved for the reasoning-heavy work.

Step 3: Re-learn the workspace — panes, not one window

This is the actual behavior change, and it's worth doing deliberately in your first session rather than drifting into it.

In Cursor, you work in one window with one model active. In meshcode, you split the workspace into panes and each pane runs its own agent, on its own model, at the same time. A natural first split for a migrating project:

<path d="M215 100 L275 100" stroke="#00ff41" stroke-width="2" marker-end="url(#arrow)"/>
<defs>
  <marker id="arrow" markerWidth="8" markerHeight="8" refX="6" refY="3" orient="auto">
    <path d="M0,0 L6,3 L0,6 Z" fill="#00ff41"/>
  </marker>
</defs>

<rect x="290" y="30" width="130" height="140" rx="8" fill="#10171e" stroke="#2b3a30"/>
<rect x="290" y="30" width="130" height="26" rx="8" fill="#0f1a13"/>
<text x="355" y="48" fill="#00ff41" text-anchor="middle" font-weight="700" font-size="12">your Claude</text>
<text x="355" y="105" fill="#7d8590" text-anchor="middle" font-size="12">migration</text>
<text x="355" y="122" fill="#7d8590" text-anchor="middle" font-size="12">refactor</text>

<rect x="430" y="30" width="130" height="140" rx="8" fill="#10171e" stroke="#2b3a30"/>
<rect x="430" y="30" width="130" height="26" rx="8" fill="#0f1a13"/>
<text x="495" y="48" fill="#00ff41" text-anchor="middle" font-weight="700" font-size="12">meshcode</text>
<text x="495" y="105" fill="#7d8590" text-anchor="middle" font-size="12">test</text>
<text x="495" y="122" fill="#7d8590" text-anchor="middle" font-size="12">scaffolding</text>

<rect x="570" y="30" width="130" height="140" rx="8" fill="#10171e" stroke="#2b3a30"/>
<rect x="570" y="30" width="130" height="26" rx="8" fill="#0f1a13"/>
<text x="635" y="48" fill="#00ff41" text-anchor="middle" font-weight="700" font-size="12">your Codex</text>
<text x="635" y="105" fill="#7d8590" text-anchor="middle" font-size="12">second</text>
<text x="635" y="122" fill="#7d8590" text-anchor="middle" font-size="12">feature</text>

A workable first-week split: your primary Claude or Codex handles the work you'd have done in Cursor's main window; a second pane on the built-in meshcode model chews through smaller mechanical tasks in parallel; a third pane stays free for a second project or a quick one-off question. You don't have to use all the panes at once — the point is that you can, whenever a task is genuinely separable.

Step 4: Adjust your habits, not your project

A few habit changes that make the first week smoother:

• Stop thinking "wait for the agent, then do the next thing." If two pieces of work don't depend on each other, put them in separate panes and let them run concurrently instead of queuing them in one thread.
• Route by cost, not by habit. Save your paid Claude/Codex pane for the reasoning-heavy work; send the repetitive stuff to the built-in model.
• Expect a native window, not a browser tab. meshcode is a native Mac/Windows app, so panes open quickly and don't carry Electron overhead — there's no tab-restore delay to get used to.
• Non-developers on your team can use meshcode too. If you've got a PM or designer who could never get comfortable in Cursor's editor-first interface, they can describe what they want in plain language in a meshcode pane.

First-hour checklist

• Install meshcode and open your existing project folder (same repo, same .git).
• Connect your Claude and/or Codex CLI subscription inside meshcode.
• Do a top-up (roughly $1 on Stripe) if you plan to use the built-in meshcode model — no monthly commitment either way.
• Open a second pane and try one small, independent task on the built-in model while your main pane keeps working.
• Keep whatever project instructions file you used in Cursor; it still works as plain-language context.
• Note which tasks felt naturally parallel — that's your cue for how to split panes going forward.

What stays exactly the same

It's worth saying plainly: your code, your repo, your git history, your deploy process, your .env files, your CI config — none of it changes. meshcode doesn't own your code and doesn't wrap it in anything. It's a native app that reads and writes the same files a terminal or Cursor would, and you can point any other tool at that same folder tomorrow if you want to. There's no export step because there was never an import step.

Cursor vs. meshcode, side by side

meshcode is in early access. Check the download page for current pricing.

Who this migration fits

• You're already comfortable in Cursor but want to stop paying a flat $20/month for weeks where you barely build.
• You have a Claude or Codex subscription sitting mostly idle outside your editor and want to actually use it inside your coding workflow.
• You regularly have two or more genuinely independent tasks (a refactor and a new feature, say) and are tired of running them one at a time.
• You want a teammate who isn't a developer to be able to contribute using plain language, without learning an editor.
• You want to confirm, in writing, that switching tools doesn't mean touching your code, your repo, or your deploy pipeline — because it doesn't.

If your workflow is deeply tied to Cursor-specific editor extensions and you ship every single day without gaps, there's less urgency to move. For anyone whose usage is bursty, whose model needs vary task-to-task, or who already pays for Claude or Codex separately, the migration above takes under an hour and the difference shows up in the first week.

👉 Download meshcode — Mac, Windows