Plotter Studio

An MCP server that gives AI agents eyes and a robotic arm.

Plotter Studio connects AI agents to an AxiDraw pen plotter and a webcam via the Model Context Protocol, enabling the agent to compose generative art as SVG, send it to the plotter, observe the result through the camera, and iterate.

How it works

1. The agent composes SVG artwork sized to your paper

2. Plotter Studio sends the SVG to the AxiDraw via the NextDraw API

3. The agent requests pen changes between passes (you do the swap)

4. The agent captures a webcam frame to see the result

5. The agent composes the next layer based on what it sees

6. Repeat until the piece is done

Related MCP server: inkscape_mcp

Requirements

• uv (Python package manager)

• Python 3.13+

• AxiDraw or NextDraw pen plotter (NextDraw firmware)

• USB webcam or gphoto2-compatible camera (e.g. Sony A7III)

• Node.js 18+ (for mcp-remote)

• gphoto2 (optional, for tethered camera backend)

Quick start

1. Clone and install

git clone https://github.com/ucodia/plotter-studio-mcp.git
cd plotter-studio-mcp
uv sync

This creates a virtual environment and installs all dependencies (FastMCP, nextdraw-api, OpenCV, Pillow).

2. Set up a camera

Plotter Studio supports two camera backends: OpenCV (USB webcams) and gphoto2 (tethered cameras like Sony, Canon, Nikon).

Option A: USB webcam (default)

Plug in your webcam, then find its device index:

uv run python -c "
import cv2
for i in range(10):
    cap = cv2.VideoCapture(i)
    if cap.isOpened():
        w, h = int(cap.get(3)), int(cap.get(4))
        print(f'  {i}: {w}x{h}')
        cap.release()
"

Set CAMERA_INDEX to the index you want to use (default is 0).

Raspberry Pi note: The pip opencv-python-headless wheel may lack V4L2 support on ARM. If the camera is detected by lsusb but OpenCV can't open it, install the system package and recreate the venv with system site-packages:

sudo apt install python3-opencv
uv venv --system-site-packages --python python3
uv sync

Option B: gphoto2 (tethered camera)

For higher quality captures, you can use a tethered camera via gphoto2. This works with many Sony, Canon, and Nikon cameras connected over USB.

Install gphoto2:

# macOS
brew install gphoto2

# Debian/Ubuntu/Raspberry Pi OS
sudo apt install gphoto2

Set your camera to PC Remote USB mode (the exact menu path varies by model), connect it via USB, and verify detection:

gphoto2 --auto-detect

Then set CAMERA_BACKEND=gphoto2 in your .env file. Captures will be slower (a few seconds per shot) but image quality will be significantly better. Set your camera to JPG-only (no RAW) to speed up transfers.

3. Run the server

uv run plotter-studio

This starts the MCP server over SSE at http://127.0.0.1:8888/sse. To listen on all interfaces (e.g. when running on a remote Pi), pass --host:

uv run plotter-studio --host

4. Configure Claude Desktop

Open your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the plotter-studio server:

{
    "mcpServers": {
        "Plotter Studio": {
            "command": "npx",
            "args": [
                "mcp-remote@latest",
                "http://127.0.0.1:8888/sse",
                "--allow-http"
            ]
        }
    }
}

5. Restart Claude Desktop

Quit and reopen Claude Desktop. You should see Plotter Studio listed as a connected MCP server (look for the hammer icon in the chat input area). If it doesn't appear, check the MCP logs in Claude Desktop's developer console.

6. Test it

Ask Claude something like:

What's the plotter status?

or

Capture a photo from the camera so I can see the paper.

If both respond without errors, you're ready to make art.

Tools

Environment variables

All configuration is via environment variables:

Notifications

Plotter Studio can send push notifications so you don't have to watch the screen while it plots. Set WEBHOOK_URL to a ntfy.sh topic URL and subscribe on your phone. You'll get pinged when a plot starts, finishes, errors, or when the agent needs you to swap a pen.

Generic JSON webhooks also work. If the URL doesn't contain "ntfy", the server POSTs a JSON body with event, timestamp, and event-specific fields.

Project structure

src/plotter_studio/
    server.py       # MCP server, tool definitions, HTTP routes, config
    plotter.py      # AxiDraw control and state machine
    camera.py       # Webcam and gphoto2 capture
    filestore.py    # Temp file store for HTTP file transfers
    webhook.py      # Push notifications
tests/
    test_plotter_state.py

License

AGPL-3.0-or-later