Claude Code for product managers: use your team's Claude Code or Cursor without a terminal

Originally published on butterneck.com.

Search for "Claude Code for product managers" or "Cursor for PMs" and most guides give similar advice: get Claude Code or Cursor (Claude Code now also runs in a desktop app and on the web at claude.ai/code, so a terminal is optional), connect the repo, learn enough git, and open pull requests yourself. For some product managers that is the right move. For many product teams it is not, and there is another way to get the same coding assistant working on a PM's request.

This post covers both. We make Butterneck, which takes the second route, so we are not neutral. We have tried to be clear about when a PM should just run Claude Code themselves.

Claude Code for product managers, in short: A PM can run Claude Code or Cursor themselves, which means repo access, a paid AI seat, usually a local setup, and opening pull requests. The other way: the PM asks in plain words on a live preview, and the Claude Code, Cursor or Antigravity CLI already signed in on the developer's computer makes the change. Butterneck does not commit, push or open a pull request; a developer decides first what ships, and Butterneck never merges or deploys.

The usual advice: give the PM their own coding assistant

The typical guide goes like this. The PM gets access to the repository, sets up Claude Code (in a terminal, an IDE, the desktop app or the web) or Cursor, runs the project, asks the assistant for a change, and then commits it and opens a pull request for a developer to review.

That works. The desktop app and the web version remove the terminal, but not the rest: the assistant still edits a copy of the code, and the PM still needs a way to run the app to see the result. With a local setup, the PM also owns the development environment: dependencies, environment variables, a dev server, branches, merge conflicts. The developer still reviews the change, but now reviews a pull request from someone who may not be able to answer questions about the diff. We looked at that trade-off in more depth in should product managers ship code?.

Cursor's own page for product managers describes this path well: PMs prototype features and ship small changes directly, starting with Cloud Agents. It works when the PM wants to drive the agent themselves. If the PM would rather ask in plain words on the running app and leave the agent, the diff and the release to a developer, that is the other way below. With Butterneck the developer's own Cursor (or Claude Code, or Antigravity) does the edit on their computer.

The other way: an AI coding agent for product managers that runs on the developer's computer

Your team probably already has a coding assistant signed in on a developer's computer. Butterneck lets the rest of the product team use it, without a terminal and without a copy of the repo.

• A developer connects a computer that already has the project and a coding CLI signed in: Claude Code (claude), Cursor's CLI (agent), or Antigravity (agy).
• The app runs on that computer's dev server, and Butterneck gives it a share link. Preview traffic is relayed through Butterneck, so nobody needs a VPN, a deploy, or localhost access.
• The PM opens the link, sees the running app, and types the request into a chat next to the screen they mean. Or someone adds the label butterneck to a Jira ticket and assigns it. See the Jira integration.
• The CLI on the developer's computer edits the connected project. The preview updates on the same link.
• A developer reads the conversation, checks the change, and publishes it the way the team already does. Or throws it away.

Butterneck never opens a pull request, never merges, and never deploys. A developer decides what goes live.

How it works

1. A developer installs the Butterneck connector on a computer with Node.js 22 or newer, a signed-in coding CLI, and the project folder.
2. They point it at the project and start the app. Butterneck creates a live preview with a share link.
3. They choose who can send requests: a list of people, or anyone with the link. People who only look need no account.
4. The PM opens the link and asks for the change in plain words, on the page it is about.
5. The coding CLI makes the edit in that folder, and the PM checks it on the preview. If it is wrong, the turn can be stopped and its file changes are usually rolled back (if not, the chat says so).
6. A developer reviews the result and ships it through the normal process: commit, pull request, CI, deploy.

PM runs Claude Code vs PM asks on a preview

When PMs should just use Claude Code themselves

Be honest about your team. Giving the PM their own coding assistant is often the better call when:

• The PM is comfortable with git and a coding assistant (terminal, desktop app or web), and wants to learn the codebase.
• They are prototyping something new, not changing a live product.
• The work is in a separate repo, a script, or an internal tool that no developer needs to approve.
• Your team already has a review process for PM-authored pull requests, and it works.
• No developer computer can stay on to run the project and the preview.

In those cases, a PM with their own Claude Code or Cursor seat is simpler than routing requests through someone else's setup.

What to know before you try this

• Your AI provider's terms apply. The coding CLI sends code context to its own provider, under that provider's terms. Butterneck does not run its own model.
• Not everything stays on the computer. Files are edited on the connected computer, but chat messages and preview traffic go through Butterneck.
• .env is a guardrail, not a hard block. Secret settings files are on an ignore list the AI is told to stay out of. That is an instruction, so keep production secrets off that computer.
• The AI stays in the folder you connected, and you can add ignore rules for paths it should not touch.
• The computer has to be on for the preview to load and for work to start.

If your team needs written rules for who can ask the AI for what, start from our AI change request policy template. For why we keep a developer at the last step, read human-in-the-loop AI coding.

What it costs

Starter is free for one developer and one person who asks for changes. Pro is USD 19 per person per month, billed monthly, with 20% off annual billing. Developers and people who request changes both count as seats; people who only open a share link to look do not. You keep paying for your own coding assistant subscription, on a plan that allows use on your team's behalf (see the note on plans and terms below). A workspace owner on Starter can also start a one-month Pro trial from the Billing page, once per workspace. Details on pricing.

FAQ

Can product managers use Claude Code without a terminal?

Yes. Claude Code also runs in Anthropic's desktop app and in the browser at claude.ai/code, so a PM doesn't strictly need a terminal. They still need a paid Claude plan, access to the repository, and usually a way to run the app to check the change. With Butterneck the PM needs none of those: they type the request in a chat on a live preview, and Claude Code runs on a developer's computer where it is already signed in.

Does a product manager need repository access to use Claude Code or Cursor?

To run it themselves, yes. Claude Code and Cursor work on code the PM can reach: a local clone, or, for Cursor's Cloud Agents, only repositories the PM's own Git account can already reach (Cursor's security docs). With Butterneck the PM doesn't need repository access. They open a share link to the app running on a developer's computer and type the request, and the Claude Code, Cursor or Antigravity CLI signed in there edits the project. The developer chooses who can send requests, and a developer decides what ships.

Does each PM need their own Claude Code or Cursor subscription?

They don't run or sign in to a coding tool themselves. Requests from the live preview run on the coding CLI already signed in on the developer's connected computer (Claude Code, Cursor's agent CLI, or Antigravity), so check that its plan allows use on your team's behalf; whether that means extra seats with your AI vendor depends on its terms (see the note below). On Pro, each person who sends requests counts as a seat; Starter includes one developer and one requester for free.

A note on plans and terms. Butterneck runs requests through the AI coding tool your developer is already signed into, so that tool's terms apply. The signed-in plan must allow use on your team's behalf. Personal and consumer plans (Claude Pro or Max, Cursor's individual plans, personal Google AI plans) may not. Anthropic's consumer terms, for example, say you "may not make your Account available to anyone else," and Claude Code's docs say Pro and Max limits "assume ordinary, individual usage." Team, Business or Enterprise plans, where each person has a seat, or an organization API key (Anthropic Console, Gemini API / Vertex AI) are the safer fit. Check your plan's terms or ask your vendor. This is not legal advice.

Does Butterneck open pull requests for the PM's change?

No. It edits files in the connected project and stops. A developer decides whether to commit it, open a pull request, or discard it. Butterneck never merges and never deploys.

Is there an AI coding agent for product managers that works on our existing codebase?

Yes. Butterneck is one option for a team that already has a web app. A developer connects a computer that already runs the product and has Claude Code, Cursor or Antigravity signed in. The product manager opens a live preview of the existing app, types the change in a chat, and checks the result on that preview. The agent edits the real project on the developer's computer. Butterneck does not commit, push or open a pull request: a developer decides first what ships, and Butterneck never merges or deploys. Tools such as Alloy, Clogless and Plusma instead connect your GitHub repository, run the change in a hosted preview and open a pull request.

Where are files edited, and what leaves the computer?

The files are edited on the connected computer. But the coding CLI sends code context to its AI provider under that provider's terms, and chat and preview traffic are relayed through Butterneck.

Is there a free plan?

Yes. Starter is free for one developer and one person who asks for changes.

Try it

If your PMs keep asking for small changes that sit in the backlog, book a 30-minute demo and we will run a real request on a live copy. Or see pricing and start on the free plan.