Skip to content
New — orchestrate a team of Mel agents from your terminal. Learn more about orchestrating Mel agents
Mel
← All articles

Mel – the agentic terminal
The agentic team for coding.

How to use Mel and Cursor together

November 20, 2026 · 7 min read

Share
In short
  • Keep Cursor for code you shape by hand; hand Mel the terminal work: builds, tests, upgrades and migrations.
  • Give Mel its own git worktree so its agents never edit the file you have open.
  • Mel reads the AGENTS.md and CLAUDE.md files you already keep, so both tools follow the same rules.
  • Mel's agent is free with your own API key on Mac, Windows and Linux.

You can use Cursor and Mel – the agentic terminal (openmel.dev) together without changing how you edit. Keep Cursor as your editor for the code you write and shape by hand. Open the same project in Mel and hand it the terminal work: running builds and tests, upgrades, migrations and longer tasks with a clear finish line. Give Mel's agent its own git worktree and it never edits the file you have open, and you review its diff before you merge. If you have the Mel API plan, Cursor's own agent can also hand a task to Mel through mel mcp serve.

This guide shows the split that works, three ways to set it up, how to keep the two from editing the same files, what each one costs, and the mistakes to avoid.

Why use an AI editor and an agentic terminal together?

An AI editor and an agentic terminal are good at different moments of the same day.

MomentBetter toolWhy
Writing a new function you want to shape line by lineCursorYou see each change as it happens
Asking about the code in front of youCursorThe file is already open
"Upgrade this library and make the tests pass"MelThe work is mostly running commands and reading their errors
Fixing a red CI buildMelThe agents run the same commands CI runs
A migration across many filesMelA lead agent can split it and check the result on disk
Running Claude Code and Codex side by sideMelEach opens as a full session in its own tab

The two do not compete for the same job. One keeps you close to the code. The other takes work off your hands and gives you back a diff to review.

What should go to Cursor and what should go to Mel?

A simple rule: if you want to watch every line, use the editor. If you want to come back to a result, use the terminal.

Send to Cursor:

  • New code you want to design yourself
  • Small edits in the file you are reading
  • Questions about a function or a file

Send to Mel:

  • Tasks that end in a passing command: "make npm test pass", "make the build green"
  • Upgrades and migrations
  • Anything you would otherwise do by typing commands, waiting, reading errors and trying again
  • Parallel work, such as a back-end change and a front-end change that do not touch the same files

Our guide on how to write a task an agent can finish shows how to phrase the second kind.

Setup 1: Mel next to Cursor (the simplest)

  1. Download Mel for Mac, Windows or Linux and open it. The terminal needs no account; Mel's AI agent needs a free account and either your own provider key or a paid plan.
  2. Open your project folder in a Mel tab with cd.
  3. Ask Mel's agent. Press Ctrl+Enter (⌘↵ on a Mac) and describe the task with a finish line, for example: "Upgrade zod to the latest version and make npm test pass."
  4. Approve what it asks. Mel reads freely, but edits and commands that change things stop at an approval card unless your settings allow them. See tools and approvals.
  5. Review the diff. Press View changes (Ctrl+Shift+=, or ⌘⇧= on a Mac) to open Mel's code review panel. The same changes also show up in any git view you already use.

This is enough for most people. The risk is that you and the agent edit the same file at the same time. Setup 2 removes it.

Setup 2: give Mel its own worktree

A git worktree is a second folder on its own branch, from the same repository. Mel makes one for you.

  1. In your project, open the + new-tab menu in Mel and choose Track this repo for worktrees.
  2. Choose Worktree: your-repo from the same menu. Mel makes a new branch, checks it out under ~/.mel/worktrees/<repo>/<branch> and opens a tab there.
  3. Give the task to Mel's agent in that tab.
  4. Keep editing in Cursor in your main folder. Nothing the agent does touches it.
  5. When the agent is done, review the branch and merge it with git.

You can open more than one worktree to run several tasks at once, or type claude or codex in a worktree tab to run those agents there. Our guide on running Claude Code and Codex together covers that.

Setup 3: let Cursor's agent hand tasks to Mel

Mel can run as a local MCP server, mel mcp serve. An app on the same computer that speaks MCP, Cursor included, starts it and gets four tools: start a task in a folder, check a task, cancel a task and list tasks.

What you need:

  1. Mel installed on the same computer.
  2. A Mel API key (it starts melk_), made on your account page.
  3. The Mel API plan on your account.
  4. Your own model key in Mel's Settings ▸ AI. These tasks run on your provider key.

Then add Mel to Cursor's MCP configuration, in ~/.cursor/mcp.json for every project or .cursor/mcp.json for one. On a Mac:

{
  "mcpServers": {
    "mel": {
      "command": "/Applications/Mel.app/Contents/MacOS/mel",
      "args": ["mcp", "serve"],
      "env": { "MEL_API_KEY": "melk_…" }
    }
  }
}

A task started this way can only read and search until you grant its folder in Mel's Settings ▸ Agent mode ▸ Other apps (MCP server) and turn on Edit files, Run commands or both. The paths for Windows and Linux are in use Mel from other apps.

Keep one set of project rules

Both tools work better when they follow the same rules. Mel reads the AGENTS.md and CLAUDE.md files in your repository as they are, then its own MEL.md for anything meant only for Mel. If AGENTS.md names a test command, Mel uses it as the check to run before it says it is done. See rules and MEL.md.

A short AGENTS.md that works well:

## Rules for this repo
- Run tests with `pnpm test`. Fix what fails before you say you're done.
- Never edit files under `vendor/`.
- Use TypeScript strict mode.

What does it cost to use both?

Mel's terminal is free. Mel's AI agent is free with your own API key, and a turn on your own key uses 0 Mel credits; your provider charges you for the usage. Paid plans add hosted credits:

Mel planPriceHosted credits each month
Free$0, with your own API keyNone
Pro$20 a month1,500
Max$200 a month18,000
Business$50 a seat a month, 2 seats minimum1,500 per seat, shared

Claude Code and Codex run on your own Anthropic and OpenAI accounts and never use Mel credits. The Mel API plan for Setup 3 is a separate plan; see pricing. Cursor's own plans are on cursor.com.

Mistakes to avoid

  • Two writers, one file. If Mel's agent and you both edit the same file, someone loses work. Use a worktree.
  • A task with no finish line. "Clean up the API" never ends. "Make test_rate_limit pass" does.
  • Skipping the diff. An agent saying it finished is not the same as the tests passing. Read the change. Our guide on keeping an AI coding agent under control helps.
  • Two sets of rules. Keep shared rules in AGENTS.md so both tools read the same thing.

Frequently asked questions

Can I use Mel and Cursor at the same time? Yes. Keep Cursor as your editor and run Mel as your terminal on the same project. Give Mel its own worktree so it never edits the file you have open.

Is Mel a replacement for Cursor? No. Mel is an agentic terminal, not a code editor. It has a built-in code review panel and editor for checking changes, and a VS Code extension, but it does not try to be your main editor.

Can Cursor's agent send tasks to Mel? Yes, through mel mcp serve, with a Mel API key, the Mel API plan and your own model key in Mel. Tasks can only read until you grant the folder.

Does Mel read my existing project rules? Yes. Mel reads AGENTS.md and CLAUDE.md as they are, then MEL.md for rules meant only for Mel.

Which models does Mel use? Claude, GPT, Gemini, Grok, DeepSeek, Qwen, Kimi, GLM and more, from 12 providers on your own key, with an Auto mode that picks a model each turn. See which AI models you can use in Mel.

Does Mel run on Windows and Linux? Yes. Mel runs on Mac, Windows and Linux.

Try it

Download Mel, open your project in a worktree tab, and give it the task you have been putting off, the one that is mostly running commands. Keep editing in Cursor while it works. For a side-by-side view of the two tools, see Mel vs Cursor and Cursor alternatives.

*Cursor is a trademark of its owner. Mel is not affiliated with or endorsed by Cursor.*

Mel – the agentic terminal · openmel.dev. The agentic team for coding.

Found this useful? Share it with your team.
Share
Lalla Asmaa Alaoui
Written by
Lalla Asmaa Alaoui
Founder & CEO of Mel

Lalla is a hands-on tech founder, agentic infrastructure architect and the vision behind Mel.

Follow on LinkedIn

Get started

Try the agentic terminal for teams.

The Mel terminal is free to download, no account needed. The AI agent is free on your own provider key — 1,500 hosted-AI credits a month on Pro.