The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Jupytercad MCP listing page.
An MCP server for JupyterCAD that allows you to control it using LLMs/natural language.
https://github.com/user-attachments/assets/7edb31b2-2c80-4096-9d9c-048ae27c54e7
Suggestions and contributions are very welcome.
The default transport mechanism is stdio. To start the server with stdio, use the following command:
To use the streamable-http transport, use this command instead:
An example using the OpenAI Agents SDK is available at examples/openai_agents_client.py. To run it, follow these steps:
Clone the repository and navigate into the directory:
Install the OpenAI Agents SDK. A Makefile target is provided for convenience:
In examples/openai_agents_client.py, update line 13 to configure a MODEL (see supported models).
Run JupyterLab from the project's root directory:
In JupyterLab, create a new "CAD file" and rename it to my_cad_design.jcad. This file path matches the default JCAD_PATH in the example, allowing you to visualise the changes made by the JupyterCAD MCP server.
(Optional) The OpenAI Agents SDK supports tracing to record events like LLM generations and tool calls. To enable it, set USE_MLFLOW_TRACING=True and run the MLflow UI:
Run the example with the default instruction, "Add a box with width/height/depth 1":
The example includes an interactive chat interface using the OpenAI Agents SDK's
REPL utility. To enable it, set USE_REPL=True.
streamable-httpTo use the streamable-http transport, first start the MCP server:
Then, run the example with the TRANSPORT variable set to "streamable-http" in the client example.
The following tools are available: