Jupyter MCP Server by Datalayer: Let Your Assistant Read, Edit and Run Notebooks

• Browses the Jupyter server's files

  Lists files and folders on the Jupyter server recursively, with a depth limit, paging and a name pattern. Use it to find a notebook before opening it.

  Try: “List the notebooks in the projects folder on my Jupyter server.”

  Files · list_files

• Lists running and available kernels

  Shows each kernel on the Jupyter server with its id, name, state and connection details, so the assistant can pick one to run code in.

  Try: “Which kernels are running on my Jupyter server?”

  Read · list_kernels

• Opens a notebook and makes it active

  Connects to a notebook by path, or creates it, and sets it as the one later cell tools work on. You can attach a kernel that is already running.

  Try: “Open analysis/sales.ipynb and make it the active notebook.”

  Runs code · use_notebook

• Lists the notebooks opened this session

  Shows the notebooks that were opened with the open tool in this session and which one is active.

  Try: “Which notebooks do we have open?”

  Read · list_notebooks

• Restarts a notebook's kernel

  Restarts the kernel behind one notebook, which clears its variables and in-memory state.

  Try: “Restart the kernel for sales.ipynb.”

  Runs code · restart_notebook

• Disconnects from a notebook

  Stops working with one notebook and releases the resources held for it. The notebook file stays on the server.

  Try: “Disconnect from sales.ipynb.”

  Write · unuse_notebook

• Reads a notebook's cells

  Returns each cell's index, source, type and execution count. A brief format gives a quick map, and a detailed format gives full content for a range of cells.

  Try: “Give me a brief overview of the active notebook.”

  Read · read_notebook

• Adds a cell at a position

  Inserts a new code or markdown cell at the position you choose in the active notebook.

  Try: “Add a markdown cell at the top that explains this analysis.”

  Write · insert_cell

• Replaces a cell's whole source

  Replaces everything in a cell with new text and returns a diff of the change. Meant for full rewrites; the targeted edit tool is safer for small changes.

  Try: “Rewrite cell 4 to use a rolling average instead.”

  Write · overwrite_cell_source

• Finds and replaces text inside a cell

  Replaces literal text in a cell, which must match exactly once unless you ask to replace all, and returns a diff. Only the matched part changes.

  Try: “In cell 3, rename the variable df to sales_df.”

  Write · edit_cell_source

• Runs a cell and returns the output

  Runs one cell of the active notebook on its kernel with a timeout, and returns the outputs, including plots as images when allowed.

  Try: “Run cell 5 and show me the chart.”

  Runs code · execute_cell

• Adds a code cell and runs it

  Inserts a new code cell at a position and runs it in one step, with a timeout, optional streaming and progress updates.

  Try: “Add a cell that prints df.describe() and run it.”

  Runs code · insert_execute_code_cell

• Reads one cell with its output

  Returns a cell's metadata and source, and its output as readable text rather than raw notebook data.

  Try: “Show cell 5 and what it printed.”

  Read · read_cell

• Deletes cells from a notebook

  Removes the cells you name from the active notebook and can return their source so you can restore them.

  Try: “Delete cells 7 to 9.”

  Delete · delete_cell

• Clears one cell's output

  Removes the saved output and execution count of a code cell without deleting the cell.

  Try: “Clear the output of cell 6.”

  Delete · clear_cell_output

• Moves a cell to a new position

  Moves one cell from one index to another and keeps its type, source and outputs. The cells in between shift to make room.

  Try: “Move cell 1 to the end of the notebook.”

  Write · move_cell

• Runs code in a kernel without saving it

  Runs code straight in the notebook's kernel, or in a kernel you name, and does not add it to the notebook. Good for magic commands, installs and quick checks of variables.

  Try: “Run %pip install seaborn in the kernel, then print df.head().”

  Runs code · execute_code

• Points the server at another Jupyter

  Connects to a different Jupyter server by URL and token while running, after checking the server's status address. If the check fails, the old connection stays.

  Try: “Connect to the Jupyter server at http://localhost:8889 with this token.”

  Network · connect_to_jupyter

Watch out for

• It runs code on your Jupyter kernel

  execute_code, execute_cell and insert_execute_code_cell run whatever the assistant writes, with the same access as your notebook (files, network, installed packages). There is no read-only mode, so approve runs in your client and use a Jupyter with only the data you can risk. github.com

• It can overwrite and delete cells

  The edit, overwrite, move and delete tools change the notebook, and the open notebook is saved as you work. Keep notebooks in git or a backup before a long session. github.com

• The README starts Jupyter on every interface

  The quick start runs JupyterLab with --ip 0.0.0.0 and a short example token. Anyone who can reach that port and has the token can run code, so use 127.0.0.1 and a long random token. github.com

• Older versions had critical security holes

  Versions through 1.0.2 had unauthenticated management routes and a DNS-rebinding flaw, and a Host-header bypass of that fix was open until 2.1.3. The current release 2.2.4 is past all three; install the latest and avoid anything older than 2.1.3. github.com

Security advisories