Bring your ChatGPT or Claude agent into Upcut. Start creating

Your agent designs. You decide what gets cut.

Upcut runs an MCP server, so ChatGPT, Claude and other agents can design and edit in your workspace — while you watch, work alongside it, and approve every toolpath.

Start creating

Why work with an agent in Upcut.

Your agent brings the intelligence. Upcut gives it a place to work — real vectors, a canvas you share, and the tools to make things.

  • Say it in plain words.

    Describe what you want, or share an image or an existing design. You don’t need to know CAD or CAM — your agent does the drawing and the setup.

  • Real, editable vectors.

    Not a picture of a design. Everything your agent makes is an editable vector with real fonts, so you can select and change any of it yourself.

  • One canvas, both of you.

    Watch your agent work in your Upcut workspace as it happens, and edit alongside it. Give feedback where you already talk to it.

  • Formulas, written for you.

    Ask for a design that resizes, and your agent writes the parameters and formulas. Change one size and the geometry follows.

  • The agent you already use.

    ChatGPT or Claude today, another tomorrow. Upcut speaks MCP, the open standard, so your work isn’t tied to one model.

  • It knows Upcut before it starts.

    Your agent loads the upcut-workspace skill — a versioned guide to how Upcut works — so it uses the editor the way it is meant to be used.

Works with the agents you already use.

Use the one you prefer today, and connect another as your needs change.

  • ChatGPT

    Add Upcut as a connector and approve access when Upcut asks.

  • Claude

    Add Upcut as a custom connector and approve access when Upcut asks.

  • Claude Code

    Run the one-line command Upcut shows you, and Upcut’s tools are in your terminal.

  • Codex and other MCP clients

    Anything that speaks MCP connects the same way — over HTTP with OAuth, or with a per-agent key.

Connected in three steps.

A few minutes, once. After that, you just ask.

  1. Step 1

    Switch your agent on in Upcut.

    Choose the workspaces it can reach, and whether it may create and edit designs or only read them.

    Settings → Agents
    Claude: On
    Access: Read · edit
  2. Step 2

    Add Upcut to your agent.

    A connector in ChatGPT, a custom connector in Claude, or one command in Claude Code.

    ChatGPT → Connector
    Claude → Custom connector
    Claude Code → One-line command
  3. Step 3

    Ask for something.

    Your agent opens the design in Upcut. Keep talking to it where you are, and watch the work in your workspace.

    “A badge sign, 24 in across.”
    → Opening it in Upcut
Rather not click through settings?Copy the onboarding prompt and paste it to your agent. It walks you through connecting, then tells you what it can do in your workspace.

You stay in charge.

Handing work to an agent only makes sense when nothing it does is out of reach. In Upcut, nothing is.

  • It reaches only the workspaces you choose.
  • It does only what you allow — read only, or create and edit.
  • Each agent has its own key and permissions. Replace the key or switch it off at any time.
  • Every save it makes is a revision under its name, so you can restore the one before any change.
  • Nothing leaves Upcut for your machine until you export it yourself.

Under the hood.

For developers, and for anyone curious what their agent is actually calling.

  • MCP

    Upcut runs an MCP server. Agents connect over HTTP and sign in with OAuth — or with a per-agent key, for local clients and bridges.

  • Tools

    Tools mirror the editor’s own operations: components, shapes and modifiers, parameters and formulas, layers. A batch of edits applies all at once or not at all. Toolpath and export tools are next.

    • upcut_create_component
    • upcut_create_shape
    • upcut_add_modifier
    • upcut_set_parameter
    • upcut_test_formula
    • upcut_apply_commands

  • Skills

    The upcut-workspace skill teaches an agent how Upcut works. Every release is versioned and never changes afterwards; agents fetch it with upcut_get_skill.

FAQs

  • Which agents work with Upcut?

    ChatGPT and Claude, including Claude Code. Upcut connects to agents over MCP, so Codex and other MCP clients connect the same way.

  • What is MCP?

    The Model Context Protocol is an open standard for giving AI agents tools. Upcut runs an MCP server, so any agent that supports MCP works in Upcut through the same tools — the editor’s own operations.

  • Does my ChatGPT or Claude plan support connectors?

    That depends on your plan with OpenAI or Anthropic. Look for connectors in ChatGPT, or custom connectors in Claude, in your agent’s settings.

  • Can I connect more than one agent?

    Yes. Switch each one on in Settings → Agents. Every agent gets its own key and its own permissions, so you can change or remove one without touching the others.

  • Can my agent change something I didn’t ask for?

    An agent reaches only the workspaces you choose, and does only what you allow — read only, or create and edit. Every save it makes is its own revision under its name, so you can restore the one before any change.

  • Will anything be cut without my approval?

    No. Toolpaths wait for your review — each operation, its cutter and its depth — and nothing leaves Upcut for your machine until you export it yourself. Simulate and check every file before you run it.

  • Do I pay for my AI agent separately?

    Your agent runs on your own ChatGPT or Claude account, billed by OpenAI or Anthropic as it is today. An Upcut plan covers Upcut.

  • Can I build my own integration?

    Yes. Connect over HTTP with OAuth, or with a per-agent key for local clients and bridges, then load the upcut-workspace skill with upcut_get_skill.

Be one of the first to try Upcut.

Bring your agent into Upcut and make it yours. Leave your email and we’ll let you in as soon as we can.

Start creating