A live Jupyter notebook server for your agent: create notebooks, insert and run cells, and execute Python 3.12 code in a real kernel. Files are discarded when the deployment stops. No API key needed.
No configuration required
Deploy and start querying — nothing to connect.
Heavy — more memory and CPU
Sessions run up to 60 minutes before the sandbox is reclaimed.
60+ AI models
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side.
How models use it and what it is built for.
The Jupyter MCP server lets a model work in a notebook the way a person does: create a notebook, add cells, run them, read the output and iterate. Hosted, each deployment comes with its own JupyterLab server and Python 3.12 kernel, so there is nothing to install.
Running code in a notebook rather than a one-shot sandbox keeps state between cells — load a dataset once, then explore it over many steps, with every step saved as a cell you can read back.
JupyterLab is protected by a random token generated for each deployment and is never exposed to the internet directly.
Concrete operations exposed as tools.
In order. Each step assumes the previous one worked.
Nothing to fill in. The deployment installs JupyterLab, starts a Python 3.12 kernel and returns a live MCP URL, usually in under a minute.
The kernel starts with a minimal environment. Ask the agent to install libraries such as pandas or matplotlib with %pip install in a cell before using them.
Describe the analysis you want. The agent creates the notebook and builds it cell by cell.
Paste any of these into Agent Studio once the server is connected.
Create a notebook, install pandas and matplotlib, load https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv and chart tips by day.
Write and run a cell that simulates 10,000 dice rolls and plots the distribution.
Read the notebook you just made and add a markdown cell summarising the findings.
Known constraints, stated plainly.
Deploy the server hosted, then watch which tools a model actually reaches for — with full JSON input and output on every call. Switch models mid-conversation to compare how each one uses the same server.
About the Jupyter MCP server.
Python 3.12.
No. The sandbox is a CPU machine sized for analysis and scripting, not training.
No. JupyterLab listens only inside the sandbox and requires a random per-deployment token. Your agent reaches it through the MCP server, not directly.