MCP server

Velmar Cut with your AI assistant

Velmar Cut has an MCP server. Connect an AI assistant on your PC, such as Claude, and it works in Velmar Cut as you would: it enters parts and stock, runs the optimisation, reads the plan and exports it. The window shows each change as it happens. Desk work on a long job can go from hours to minutes.

7 days, every edition unlocked · Windows 10 and 11 · no account Linux and macOS: coming soon

How to use Velmar Cut with MCP

This video shows how to connect an AI app and then do a job shop's day of work by asking, with each problem noticed and solved in the chat. The chat window is an illustration and its words are written for the video, while each step in it really ran in Velmar Cut.

Connect an AI app, then do the day's work by asking · 7 min

Chapters

  1. 0:06 Connect
  2. 1:18 Pipes, by asking
  3. 2:42 Plates
  4. 4:39 The warehouse
  5. 5:40 One project
Text of the video

An AI app on this computer can work in Velmar Cut through MCP. It starts on the MCP settings page.

Start server, and the page shows it running, with its port.

Only programs on this computer can connect. There is no account and no token. Start with Velmar Cut keeps it on.

Which app do you use? Each app in this list has its own steps here.

Claude Code takes one command in a terminal, ready to copy.

Claude Desktop has an extension: Install for Claude Desktop. It needs nothing else.

The other way is Add automatically: it adds Velmar Cut's entry to the app's settings file, and keeps a copy of the file.

The ChatGPT desktop app and Codex, VS Code with Copilot, Cursor, and any other MCP app: the steps for each are here too.

Connected. The app now has 71 tools for Velmar Cut, and the page counts it as a client.

From here on nothing is clicked by hand. The chat window is only an illustration; each step in it really ran in Velmar Cut.

Velmar Cut's window, beside the chat, shows every change as it happens.

A list in the chat becomes a job, with its parts and its stock.

First problem: the office typed one tube differently from the stock. Velmar Cut notices, and asks.

A question from Velmar Cut goes to you. The AI may explain it, and Velmar Cut tells it not to answer for you.

Cancel, three rows corrected, and planned again: every piece placed.

The plan as a picture in the chat, and in the window beside it.

Second problem: a part changes after the plan, so the plan is out of date.

Planned again. Writing over the first PDF is a question for you, too.

Going over to 2D leaves a job with parts in it, so Velmar Cut asks first.

The DXF and DWG drawings come in with their outlines and holes.

Third problem: the table has a bad row. Velmar Cut refuses the file and says which row.

No drawing for a part? Say what it is. Velmar Cut draws it and shows it in the chat. For some small workshops, CAD may not even be required.

A part from a drawing can be changed the same way, on its own row.

A quick plan first, at the Fast level, to see whether everything fits.

Fourth problem: one part fits no sheet. The plan says which, and why.

Taken from the warehouse, and planned at the Balanced level. The AI follows the search while it runs.

The wall plate and the flange come out on the sheets, as drawn.

Close up, in the window: the wall plate you described, with its four holes and its slot.

And the flange, with the four new Ø8 holes around its bore.

A new order. It takes its stock from the warehouse.

Fifth problem: more is asked for than the warehouse holds. Velmar Cut refuses and says how many there are.

Sixth problem: pieces are left out of the plan. The AI asks what to do, and does not decide for you.

Booking out: the warehouse loses what the plan uses, and you choose which remnants it keeps.

Two jobs of one order go in one project file.

Optimize all jobs plans them in turn, while the window shows each one.

Booking out a whole order is a question for you again.

That's the same day's work as in the first video, this time by asking, with every problem noticed and solved in the chat.

Asking can save even more time, and leave more of the day for production. It's all inside.

Desk work from hours to minutes

The slow part of a cutting job is often the desk work around it. Hand it to an AI app over MCP, and entering a long parts list, setting up stock, planning and exporting can go from hours to minutes, and the time saved can go to production.

Entering parts and stock

You hand over a list, a spreadsheet or your drawings. The app enters the parts and stock and imports the drawings and tables, and the window shows each change.

Planning

You say which level and cutting rules you want. The app sets them, runs Optimize, follows it to the end and reads the plan, so you can ask about waste or the number of sheets, and you check the plan before cutting.

Exports and the warehouse

You say what the shop floor needs. The app exports the plan as PDF, DXF or CSV and, once you have checked it, applies the plan to the warehouse.

From a sketch to a part

Photograph a sketch of a base plate or bracket and ask your AI app: it reads the sizes, asks for missing ones and has Velmar Cut draw the part, with a picture in the chat to check. The part goes into the job, and Velmar Cut writes the DXF, DWG or SVG.

  1. Photo of the sketch
  2. Drawn in Velmar Cut
  3. DWG for the machine

What the assistant can do

MCP, the Model Context Protocol, is the open standard AI apps use to work with other programs. Through it the assistant uses Velmar Cut's own commands in the open window, so you see what it does.

Jobs

Create and open jobs and projects, for bars and profiles or for sheets.

Parts and stock

Enter parts and stock, for example from a list you give the assistant.

Levels and settings

Choose the level, from Fast to Maximum, and the cutting settings.

Optimise, read and show the plan

Run Optimize, then read the plan and answer your questions about it. Show it as pictures in the chat, and the best plan so far while Optimize runs.

Every export

Use every export in the program: PDF reports, HTML, DXF, DWG, SVG and pictures, the cut sequence, quoting data, and the parts, stock and warehouse lists.

Warehouse

Look up and use the stock and remnants in the warehouse.

Where Velmar Cut would ask you, for example to confirm a change or for the bar length of automatic stock, it puts the same question to the assistant and waits for the answer. The assistant is told to ask you and not to choose for you.

A laser job planned from the chat

The assistant sets up the job and starts the optimisation, Velmar Cut's window shows the plan, and the DXF for the laser ends up in the job folder. At the end the plan also shows as a picture in the chat.

Your AI assistant Velmar Cut connected

Import the DXF drawings in C:\Jobs\Hall 3\Laser with the quantities below and nest them on our 2500 × 1250 mm S355 sheets, 8 mm thick, at €265 a sheet. Balanced level, least waste. Then save the DXF for the laser and show me the plan.

Sprocket_z323 Sprocket_z218 Flange_DN1006 … and 12 more

I'll set up a new 2D job in Velmar Cut with the drawings and your sheets, then start the optimisation at the Balanced level.

Velmar Cut new_job Velmar Cut asks before switching to 2D

Sent

{ "mode": "2D", "name": "Hall 3 · laser S355 8 mm" }

Returned

{
  "status": "question",
  "question": { "question_id": "q1", "title": "Change cutting mode",
                "text": "Switch to 2D sheet nesting? A new job starts; the current job stays in History and Recent jobs.",
                "choices": [ "Switch to 2D", "Cancel" ] },
  "instructions": "Velmar Cut is waiting for an answer, as it waits at its own dialog. Ask the user this question with its choices … Never choose for the user."
}

Velmar Cut has a question before it starts the 2D job. Tell me your choice and I'll pass it on.

Change cutting mode

Switch to 2D sheet nesting? A new job starts; the current job stays in History and Recent jobs.

Switch to 2DCancel

Switch to 2D

Velmar Cut answer_question New 2D job “Hall 3 · laser S355 8 mm”

Sent

{ "question_id": "q1", "answer": "Switch to 2D" }

Returned

{
  "created": true,
  "job": { "name": "Hall 3 · laser S355 8 mm", "mode": "2D", "unit": "mm", "currency": "EUR",
           "part_rows": 0, "stock_rows": 0, "cutting_method": "guillotine", … },
  "answered": "Switch to 2D"
}
Velmar Cut import_cad 15 parts from the drawings, quantities added for 160 pieces in all

Sent

{
  "paths": [ "C:\\Jobs\\Hall 3\\Laser\\Sprocket_z32.dxf",
             "C:\\Jobs\\Hall 3\\Laser\\Sprocket_z21.dxf", … ],
  "into": "parts", "material": "S355", "thickness": 8,
  "rows": [ { "number": 1, "quantity": 3 }, { "number": 2, "quantity": 8 }, … ]
}

Returned

{
  "added": [
    { "number": 1, "label": "Sprocket_z32 · CAD_OUTER_CONTOURS · 49 [bounding blank]",
      "length": 204.0547, "width": 204.0547, "quantity": 3, "holes": 7, … },
    { "number": 2, "label": "Sprocket_z21 · CAD_OUTER_CONTOURS · 49 [bounding blank]",
      "length": 137.5764, "width": 137.2001, "quantity": 8, "holes": 6, … },
    …
  ]
}
Velmar Cut add_stock Sheets 2500 × 1250 mm, S355, 8 mm

Sent

{ "stock": [ { "length": 2500, "width": 1250, "material": "S355",
               "thickness": 8, "unlimited": true, "price": 265 } ] }

Returned

{ "added": [ { "number": 1, "material": "S355", "length": 2500, "width": 1250,
               "unlimited": true, "price": 265, "thickness": 8, … } ] }
Velmar Cut set_settings True-shape nesting, Balanced level, least waste

Sent

{ "level": "balanced", "objective": "waste", "cutting_method": "true_shape" }

Returned

{ "mode": "2D", "unit": "mm", "level": "balanced", "search_seconds": 300, "objective": "waste",
  "cutting_method": "true_shape", "true_shape": { "applies": true, … }, … }
Velmar Cut optimize Search started, up to 5 minutes

Sent

{ }

Returned

{
  "started": true,
  "optimization": { "running": true, "elapsed_seconds": 0, "stop_keeps_plan": false, "limit_seconds": 300, … }
}

The search is running in Velmar Cut, and its window shows the plan as it improves. I'll check back until it is done.

Velmar Cut optimization_status Plan so far: 160 of 160 pieces on one sheet

Sent

{ "wait_seconds": 40 }

Returned

{
  "optimization": {
    "running": true, "elapsed_seconds": 40.2, "stop_keeps_plan": true, "limit_seconds": 300,
    "best_so_far": { "layouts": 1, "yield_percent": 63.49, "pieces_placed": 160,
                     "pieces_total": 160, "found_after_seconds": 39.8 }, …
  }
}
Velmar Cut optimization_status Search finished after 8 status checks

Sent

{ "wait_seconds": 40 }

Returned

{
  "optimization": { "running": false },
  "plan": { "layouts": 1, "yield_percent": 63.49, "pieces_placed": 160, "pieces_total": 160,
            "cost": 265, "currency": "EUR", "calculated_in_seconds": 300.7 }
}
Velmar Cut export_plan DXF saved

Sent

{ "format": "dxf", "path": "C:\\Jobs\\Hall 3\\Laser" }

Returned

{ "exported": "C:\\Jobs\\Hall 3\\Laser\\Hall_3_·_laser_S355_8_mm__2D__1-sheets.dxf", … }
Velmar Cut show_plan The plan as a picture: one sheet, 63.5 % yield

Sent

{ }

Returned

{
  "layouts": 1, "yield_percent": 63.49, "pieces_placed": 160, "pieces_total": 160,
  "cost": 265, "currency": "EUR", "calculated_in_seconds": 300.7,
  "shown": "plan", "pictures_shown": 1
}

All 160 pieces fit on one sheet at 63.5 % yield, with a material cost of €265. The DXF for the laser is in C:\Jobs\Hall 3\Laser. The plan is open in Velmar Cut, so you can check it before cutting.

Hall_3_·_laser_S355_8_mm__2D__1-sheets.dxf

In Velmar Cut

Meanwhile Velmar Cut's window shows the plan: 160 pieces on one 2500 × 1250 mm sheet, 63.5 % yield.

A conversation from a run of Velmar Cut 1.7.0. The calls and answers are Velmar Cut's own, shortened where they are long; each AI app shows them in its own way.

Set it up

MCP settings shows everything an AI app needs and can add Velmar Cut to the app for you. Installing the Claude Desktop extension or adding an app starts the server and turns on Start with Velmar Cut, so there is nothing else to switch on.

  1. Open MCP settings

    In Velmar Cut, open MCP settings in the sidebar, directly under General settings. The top of the MCP server card says Server running · port 48356 or Server stopped, and a green dot on the sidebar item shows that the server runs.

  2. The address

    Velmar Cut takes port 48356 when it is free, so the Address is the one below. If another program holds that port, Velmar Cut takes a free one and tells you; a port saved earlier stays. Copy address copies it.

    http://127.0.0.1:48356/mcp
  3. Choose your app

    Under Connect an AI app, pick your app in Which app do you use?. Velmar Cut shows how to connect it and the entry it needs, with Copy. For Claude Desktop, Install for Claude Desktop installs Velmar Cut's extension.

  4. Add it to the app

    Add automatically writes Velmar Cut's entry into the app's settings file. It asks first, changes only that entry and keeps a copy of the file as it was. Open the file opens it if you prefer to do it yourself.

Connect your AI app

Velmar Cut's server speaks MCP over Streamable HTTP at the address it shows, and MCP settings gives the entry for each app below. The examples use the default port 48356; if MCP settings shows another port, use that one. If a step differs in a newer version of your app, its own MCP help has the current one.

Claude Code

Run this command once in a terminal (PowerShell or Command Prompt). Velmar Cut's tools are then in every Claude Code session, and /mcp shows the server.

claude mcp add --transport http --scope user velmar-cut http://127.0.0.1:48356/mcp

Claude Desktop

Install for Claude Desktop in MCP settings opens VelmarCut-Claude.mcpb, the extension that ships beside VelmarCut.exe, in Claude Desktop. If it shows the file instead, drag it into the Claude Desktop window, or pick it in Settings ▸ Extensions ▸ Advanced settings ▸ Install Extension…. The extension finds Velmar Cut by itself; its Port setting is only for a second Velmar Cut, or when another program holds port 48356. Connectors added under Customize ▸ Connectors run from Anthropic's cloud and cannot reach your PC.

Without the extension: Add automatically puts this entry in claude_desktop_config.json (Settings ▸ Developer ▸ Edit Config opens it). Then quit Claude Desktop from the system tray and start it again. This route needs Node.js 18 or later.

{
  "mcpServers": {
    "velmar-cut": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://127.0.0.1:48356/mcp", "--allow-http", "--transport", "http-only"]
    }
  }
}

ChatGPT desktop app and Codex

The ChatGPT desktop app, the Codex CLI and the Codex IDE extension share the settings file %USERPROFILE%\.codex\config.toml, so Add automatically sets up all three with this entry. In the ChatGPT app itself: Settings ▸ MCP servers ▸ Add server, the name velmar-cut, Streamable HTTP and the address, then Restart. ChatGPT on the web cannot connect.

[mcp_servers.velmar-cut]
url = "http://127.0.0.1:48356/mcp"

VS Code (GitHub Copilot)

GitHub Copilot's agent mode uses Velmar Cut once this entry is in your user mcp.json (the command MCP: Open User Configuration opens it). Add automatically puts it there; then start velmar-cut in VS Code's list of MCP servers.

{
  "servers": {
    "velmar-cut": { "type": "http", "url": "http://127.0.0.1:48356/mcp" }
  }
}

Cursor

Cursor reads this entry from mcp.json in the .cursor folder of your user profile. Add automatically puts it there, and Cursor Settings ▸ MCP then lists Velmar Cut.

{
  "mcpServers": {
    "velmar-cut": { "url": "http://127.0.0.1:48356/mcp" }
  }
}

Another MCP app

Any app that speaks MCP over Streamable HTTP connects with this address. It needs no sign-in or token: only programs on your PC can connect.

http://127.0.0.1:48356/mcp

Where an app keeps its MCP servers in a settings file, Add automatically writes the entry for you: it asks first, changes only Velmar Cut's entry and keeps a copy of the file as it was. A file with comments is left as it is, and you paste the entry in yourself.

AI agents find their own way in

Every Velmar Cut build ships AGENTS.md beside VelmarCut.exe, in the portable ZIP and in the installed folder.

Point Codex, Claude Code or another AI agent at the Velmar Cut folder, and it reads there how to connect and how to work in Velmar Cut. Show AGENTS.md on the AI agents card in MCP settings opens the file.

AGENTS.md names the default port, 48356. While the server runs, Velmar Cut also writes its address to mcp-server.json in its data folder, so an agent finds the server without your help.

Velmar Cut\
  VelmarCut.exe
  AGENTS.md
  VelmarCut-Claude.mcpb

%LOCALAPPDATA%\VelmarCut\
  mcp-server.json

On your PC, and only while it is on

  • It listens only on your own PC (127.0.0.1), never on the network.
  • It refuses requests from web pages, so a website open in your browser cannot reach it.
  • It is off until you start it or connect an AI app in MCP settings. Stop server ends it at any time, and with Start with Velmar Cut off it stays off the next time.
  • It has no password: while it is on, any program on your PC can use it. Turn it off when you do not need it.

What your AI app sees

The assistant reads what it works with, such as part lists, stock and plans. Your AI app sends that to its provider as part of the conversation, under that provider's terms, as it does with anything else you share in a chat. With a model that runs on your own PC, nothing goes to a provider.

Questions about the MCP server

What is MCP?

The Model Context Protocol is an open standard for connecting AI apps to other programs. A program offers tools, and the AI app uses them when you ask for something. Velmar Cut offers its own commands as such tools.

Which AI apps work with it?

Claude Code, Claude Desktop, the ChatGPT desktop app, Codex, VS Code with GitHub Copilot and Cursor: MCP settings shows how to connect each of them. Other apps work if they support MCP servers over Streamable HTTP on your own PC.

Does it work with ChatGPT?

Yes, the ChatGPT desktop app. Choose it in Which app do you use?, and Add automatically writes the entry to the settings file it shares with Codex. In the app: Settings ▸ MCP servers ▸ Add server, then restart it. ChatGPT on the web cannot connect to Velmar Cut on your PC.

Which language does the assistant use?

Velmar Cut describes its tools to the assistant in English. The assistant answers you in your own language.

Can the assistant change my jobs and stock?

Yes. It can do what you can do in the window, and you see each change as it happens. Where Velmar Cut asks before a change, the assistant gets the same question.

Does Velmar Cut send anything to the internet for this?

No. The MCP server only answers requests from programs on your PC. What leaves your PC is up to your AI app and its provider.

How much time does it save?

It depends on the job, and the most is saved on a long parts list: entering it can go from hours to minutes. The search itself takes as long as the level you choose, from up to 1 minute at Fast to 30 minutes at Maximum, and you check the plan before cutting.

Free trial

Seven days with every edition

Install Velmar Cut, enter the parts and stock of a current job and print the cutting schedule. The trial starts on first run and needs no account or card.